Health Insurance Status, Medical Debt, and Their Impact on Access to Care in Arizona

Abstract

Objectives. We examined the impact of health insurance status on medical debt among Arizona residents and the impact of both of these factors on access to care.

Methods. We estimated logistic regression models for medical debt (problems paying and currently paying medical bills) and access to care (medical care and medications delayed or missed because of cost or lack of insurance).

Results. Insured status did not predict medical debt after control for health status, income, age, and household characteristics. Insured status (adjusted odds ratio [AOR] = 0.32), problems paying medical bills (AOR = 4.96), and currently paying off medical bills (AOR = 3.04) were all independent predictors of delayed medical care, but only problems paying (AOR = 6.16) and currently paying (AOR = 3.68) medical bills predicted delayed medications. Inconsistent coverage, however, was a strong predictor of problems paying bills, and both of these factors led to delays in medical care and medications.

Conclusions. At least in Arizona, health insurance does not protect individuals from medical debt, and medical debt and lack of insurance coverage both predict reduced access to care. These results may represent a troubling message for US health care in general.

Financial hardship caused by out-of-pocket medical expenses is a large and growing problem in the United States. In 2007, it was estimated that 19% of the population had problems paying medical bills (up from 15% in 2003).¹ Another 2007 study estimated that 41% of the population aged 19 to 64 years had problems paying medical bills, had accrued medical debt, or both.^2,3 That estimate was up from 34% in 2005.

Health insurance coverage should protect individuals from medical debt. However, in contrast with most insurance products, which indemnify losses stemming from low-probability and high-consequence events, health insurance plays the dual role of promoting actions to prevent such events and protecting against loss from them. Furthermore, in the United States health insurance rarely provides complete protection against financial loss caused by illness or injury. Rather, increasing portions of medical costs are paid directly by the insured in the form of copayments, deductibles, exclusions and limits on covered benefits, coinsurance provisions, and lifetime spending caps. In fact, a substantial portion of individuals both insured and uninsured, as well as individuals across a wide range of ages and income levels, have reported that paying medical bills is a problem.^1–6

At least to some extent, medical debt is also more damaging than are other types of consumer debt because medical bills are often incurred through an illness or injury that limits one's ability to work. Specifically, problems paying medical costs are associated with higher credit card debt,⁷ more calls from bill collectors,⁷ increased bankruptcy rates,^7–10 and diminished access to care.^{1,4–6,11,12}

We took advantage of a large health-related data set to examine medical debt and its consequences in Arizona. Because substantial national attention has been devoted to reducing the ranks of the uninsured in an attempt to improve access to care,^13–16 we particularly focused on estimating the impact of health insurance status on the prevalence of debt and the impact of both of these factors on access to care.

METHODS

The 2008 Arizona Health Survey (AHS) was a comprehensive survey of 4200 Arizona households designed to assess health insurance coverage, health status and behaviors, and social and environmental factors that affect population health.¹⁷ Because individuals aged 65 years and older have access to Medicare, we focused on adults aged 18 to 64 years. We specified and estimated 2 sets of logistic regression models. The first set examined predictors of medical debt, including insurance status. The second set examined the relative impact of medical debt and insurance status on whether medical care or prescribed medications were delayed or forgone.

Measures

Medical debt.

Two items from the 2008 AHS (which were similar to a pair of items used in 2 national surveys^1,2) were used in constructing the medical debt variables for this study. Individuals were defined as having problems paying bills if they answered yes to the following question: “During the past 12 months, were you unable to pay or did you have problems paying for medical bills, either for yourself or any family member in your household?” Individuals were defined as having current medical bills if they answered yes to the question “Are you currently paying off any medical bills?” (The term “medical debt” refers to a yes answer to either of these items.) The medical debt variables were the dependent variables for the first set of models and were explanatory variables in the second set.

Access to care.

We measured access to care (the dependent variable for the second set of models) using 4 items from the 2008 AHS: 2 items focusing on access to medications and 2 items focusing on access to medical care. In each case, the first item in the pair asked whether, in the preceding 12 months, the respondent had delayed or not obtained a medicine that was prescribed (for access to medications) or had delayed or not obtained any other needed medical care (for access to medical care).

Because a person can delay or not obtain care for a number of reasons (e.g., a medication may be delayed because of lack of time to pick it up), individuals were identified as having a problem with access only if they also answered yes to a second item asking whether cost or lack of insurance was a reason. (We refer to a delay in obtaining or not obtaining a prescribed medicine and medical care because of cost or lack of insurance as “delayed medications” and “delayed care,” respectively.)

Insurance.

Several insurance variables were included as independent variables in both models. These included whether individuals had health care coverage (insured status), whether they were insured at the time of the survey but had been uninsured at some point during the preceding year (inconsistent coverage), extent of coverage (limited benefits are one measure of underinsurance^18–20), and type of insurance (e.g., Medicaid, employer-sponsored insurance).

Health status.

Health status was also included in both models. An individual's health determines whether medical costs are incurred and can also influence whether care or medications are delayed (e.g., because poor health can strain financial resources).²¹

Health status variables included reports of chronic conditions (whether an individual reported a chronic physical condition, psychological condition, or both) and self-reported ratings of general health (fair or poor vs good or excellent). Individuals who reported being told by a doctor that they had asthma, diabetes, hypertension, heart disease, arthritis (or gout, lupus, or fibromyalgia), or gastrointestinal (stomach or digestive) disorders were defined as having a chronic physical condition. Similarly, individuals who reported being told by a doctor that they had bipolar or manic–depressive disorder, anxiety disorder, or major or clinical depression were defined as having a chronic psychological condition.

Income.

Individuals who have sufficient incomes (and presumably more financial reserves) are less likely than are those who do not to have problems paying medical bills or to be currently paying off medical bills. They are also less likely to have problems with access to care and to report cost as a reason. Therefore, we included household income as an independent variable in both sets of models.

Data on income were missing for 13% of the sample. To maximize the power of our models, we estimated missing income data in a separate analysis with the following variables as predictors: 2000 US census tract median household income,²² age, education, whether a household was headed by a couple or a single individual, type of home and whether it was owned or rented, whether the household was receiving state financial assistance, and whether the respondent indicated that health services were delayed because of cost. The R² value of the estimating model was 0.59. To assess whether those with missing reported income data (for whom estimated income was used) were inherently different with respect to their likelihood of medical debt or access problems, we included a variable indicating whether income was reported or estimated.

Ethnicity.

Because there is evidence that the prevalence of medical debt is higher in minority populations,⁴ and because it has been argued that differences in access to care are a core reason for health disparities,²³ the final variable included in both models was whether individuals identified themselves as Hispanic or Latino.

Additional medical debt independent variables.

We included several independent variables in the medical debt models only. Because individuals in the lowest income categories have access to more financial assistance programs (i.e., those with incomes below the federal poverty level are eligible for Arizona's Medicaid program), we included Medicaid eligibility in our models as an additional income variable.

In addition, we included age and age squared to allow for the possibility of an inverted U-shaped relationship between medical debt and age. This shape is consistent with the hypothesis of a lower likelihood of medical debt in the young (as a result of good health) and among those nearing the age of 65 years (as a result of longer workforce involvement and its attendant higher likelihood of adequate health insurance coverage and financial reserves) than among those in middle age. Finally, we included household status (whether the household was headed by a couple or a single individual and whether children were present) to capture the number of people who might be contributing to and dependent on the household's income and who might have health problems that could result in medical debt.

Additional access to care independent variables.

Similarly, we included several independent variables in the access to care models only. Individuals who report that they have a usual source of health care are believed to be generally more engaged in their health and less likely to delay care than those without a usual source of care.^24–26 They may also be less likely to delay taking medications because their prescribing physician monitors and encourages medication adherence.²⁷

Another possible element in delay of care or medication is its perceived value. If the care or medication is considered to be more worthwhile, it is less likely to be delayed or missed even when cost or insurance is an issue. The perceived value of care or medication may increase with age because of the increased responsibilities and deteriorating health faced by older adults. Perceived value may also differ according to gender (as a result of gender-related perceptions of health) and marital status (as a result of the existence of another person interested in an individual's health).^28–31

Statistical Analysis

We used weighted data to estimate prevalence rates and logistic regression (SPSS version 17.0.0, SPSS Inc, Chicago, IL) and unweighted data to estimate the models. Blocks of explanatory variables were entered into each set of models sequentially according to a defined causal theory.³² The work by Andersen et al. work provided the basis for the access models.³³ Each block contained a set of related variables believed to define the causal mechanism in question. The blocks were ordered from those hypothesized to be most proximal to those most distal in causing medical debt (first 2 models) and delayed care or medication (second 2 models).

The single exception to this scheme is that medical debt was included last in the delayed care and medication models to determine whether it was explanatory even after all other variables had been taken into account. Sequential entry of these blocks allowed the incremental explanatory power of each block to be measured and tested in a logical order. The adjusted odds ratios (AORs) reported were derived from the final models unless otherwise indicated. We report incremental results from the sequential entry of blocks of variables only when they differed in statistical significance from what was found in the final models.

RESULTS

Complete data on 2368 individuals (89% of the sample) were available for our analysis. Demographic data on the sample are provided in the appendix available as a supplement to the online version of this article at http://www.ajph.org.

Medical Debt

Across all insured status categories, reports of problems paying medical bills were more prevalent than were reports of paying off medical bills (Table 1). Problems paying medical bills and medical debt in general were more prevalent among those without insurance. All measures of medical debt were highest among those who were insured at the time of the survey but had been uninsured during some portion of the preceding year (inconsistent coverage). The frequency of delayed medications and care was highest among those who reported problems paying medical bills and lowest among those who reported no medical debt problems (Table 2).

TABLE 1.

Report of Each Type of Medical Debt, by Insured Status (Weighted Sample): Arizona Residents Aged 18–64 Years, 2008

	All Persons in Arizona Aged 18–64 Years, %	Problems Paying or Unable to Pay Medical Bills in Past Year, %	Currently Paying Off Medical Bills, %	Problems Paying or Currently Paying Medical Bills, %	No Medical Debt, %
All persons in Arizona, 18–64 y	100	20	19	29	71
No insurance coveragea	17	28b	19	36b	64b
Insured	83	18c	19	27c	73c
Currently insured but uninsured in past y	7	37	21	43	57

^aMedical debt prevalence not statistically different between those with no insurance and those currently insured but uninsured in the past year.

^bMedical debt prevalence statistically different than for those with insurance coverage at P < .05.

^cMedical debt prevalence statistically different than for those currently insured but uninsured in the past year at P < .05.

TABLE 2.

Reports of Delayed Medications and Care, by Type of Medical Debt (Weighted Sample): Arizona Residents Aged 18–64 Years, 2008

	All Persons in Arizona Aged 18–64 Years, %	Problems Paying or Unable to Pay Medical Bills in Past Year, %	Currently Paying Off Medical Bills, %	Problems Paying or Currently Paying Medical Bills, %	No Medical Debt, %
Delayed or did not obtain a prescription	16	39a	32a	33a	9
Delayed or did not obtain a prescription, with insurance or cost one reason	9	31a	24a	25a	3
Delayed or did not obtain needed care	20	45b	39b	39b	13
Delayed or did not obtain needed care, with insurance or cost one reason	13	39b	32b	32b	5
No delayed care or prescriptions	69	40	49	47	79
No delayed care or prescriptions with insurance or cost as a reason	82	51	60	59	92

^aDelayed prescription prevalence statistically different (P < .05) for those with this type of medical debt than for those with no medical debt.

^bDelayed care prevalence statistically different (P < .05) for those with this type of medical debt than for those with no medical debt.

Logistic regression model estimates of problems paying medical bills showed that although health insurance coverage was predictive when considered alone (OR = 0.53; 95% confidence interval [CI] = 0.40, 0.71; data not shown), it was no longer predictive when the other variables were included in the model (Table 3). In fact, the only insurance-related variables that remained statistically significant in the full model for problems paying bills were whether individuals had consistent coverage and whether they had Medicare coverage only. Those with inconsistent coverage had almost 2.5 times the odds of those with continuous insurance to have problems paying medical bills (AOR = 2.48). Individuals aged 18 to 64 years who had only Medicare coverage had twice the odds of those with only employer-sponsored insurance to have problems paying bills (AOR = 1.99).

TABLE 3.

Estimated Adjusted Odds Ratios (AORs) for Models Predicting Problems Paying Medical Bills and Currently Paying Off Medical Bills: Arizona Residents Aged 18–64 Years, 2008

Variable	Problems Paying Bills, AOR (95% CI)	Paying Off Bills, AOR (95% CI)
Insured	0.88 (0.49, 1.57)	1.15 (0.64, 2.06)
Inconsistent coverage	2.48 (1.61, 3.82)	1.22 (0.77, 1.93)
Underinsurance
Prescriptions	0.76 (0.45, 1.26)	0.69 (0.42, 1.14)
Mental health	0.83 (0.61, 1.13)	1.15 (0.86, 1.54)
Dental	1.00 (0.74, 1.36)	1.28 (0.95, 1.72)
Health status
Chronic physical condition	1.79 (1.33, 2.41)	1.45 (1.11, 1.89)
Chronic psychological condition	2.53 (1.47, 4.33)	1.14 (0.64, 2.01)
Both	0.73 (0.40, 1.35)	1.39 (0.74, 2.63)
Fair or poor health	1.96 (1.47, 2.61)	1.72 (1.29, 2.28)
Income
Household income	0.99 (0.982, 0.990)	0.99 (0.991, 0.996)
Medicaid eligible	0.55 (0.38, 0.82)	0.50 (0.33, 0.75)
Income data missing	0.89 (0.30, 2.63)	0.23 (0.05, 0.96)
Type of insurance (Ref: employer-sponsored insurance only)
Medicare only	1.99 (1.12, 3.54)	2.39 (1.37, 4.19)
Medicaid only	1.15 (0.73, 1.82)	0.82 (0.50, 1.32)
Direct purchase only	1.60 (0.93, 2.76)	1.41 (0.87, 2.30)
Other/combination	1.06 (0.66, 1.71)	1.10 (0.69, 1.74)
Age	1.09 (1.02, 1.18)	1.11 (1.04, 1.20)
Age squared	0.999 (0.998, 0.999)	0.999 (0.998, 0.999)
Household status (Ref: couple as head of household)
Single female head of household	0.91 (0.62, 1.33)	0.94 (0.66, 1.32)
Single male head of household	0.67 (0.42, 1.08)	0.59 (0.37, 0.93)
Children in household	1.62 (1.14, 2.31)	1.62 (1.18, 2.21)
Single woman and children	1.31 (0.77, 2.24)	1.06 (0.64, 1.77)
Single man and children	0.73 (0.29, 1.86)	0.97 (0.40, 2.34)
Hispanic	1.03 (0.76, 1.39)	0.78 (0.58, 1.05)
Nagelkerke R2	0.209	0.118

Note. CI = confidence interval.

Health insurance status had even less of an effect on whether individuals were currently paying off medical bills. Whether someone had health insurance was not predictive when entered alone in the model (OR = 0.98; 95% CI = 0.72, 1.33; data not shown), nor was it statistically significant when all of the variables were entered (Table 3). The only statistically significant insurance-related variable with respect to current medical bills indicated that nonelderly adults with Medicare coverage only were more likely than were those with employer-sponsored insurance only (AOR = 2.39) to be currently paying off medical bills.

Health status tended to be a significant predictor of medical debt of both types, although the primary impact of reporting a chronic psychological condition seemed to be on problems with paying medical bills. People diagnosed with a chronic psychological condition were more than 2.5 times as likely as were those without such diagnoses (AOR = 2.53) to have problems paying medical bills. As expected, the odds of medical debt decreased as income increased (AOR = 0.99). However, people with income levels low enough for Medicaid eligibility had roughly half the odds of both types of medical debt of those with higher incomes. Also as expected, the odds of medical debt increased with age (AORs = 1.09 for problems paying medical bills and 1.11 for currently paying off bills) but at a decreasing rate (age squared AOR = 0.999). Finally, the odds of medical debt increased when there were children in the household (AOR = 1.62 for both models).

Delayed Care and Medications

Although insured status was not predictive of medical debt, it appeared to have an impact on delayed care (Table 4). All insurance blocks explained a statistically significant amount of variance when they were added in sequence to the delayed medications model, and all but the underinsurance block were statistically significant with respect to delayed care (data not shown). However, after all of the variables had been included simply having insurance coverage predicted delayed care only. People with insurance had one third the odds of delayed care of those without insurance (AOR = 0.32). By contrast, inconsistent coverage was associated with 6 times the odds of delayed care (AOR = 5.99) and almost 5 times the odds of delayed medications (AOR = 4.67), even after all of the explanatory variables had been controlled.

TABLE 4.

Estimated Adjusted Odds Ratios (AORs) for Models Predicting Delays in Obtaining or Inability to Obtain Needed Care and Medications: Arizona Residents Aged 18–64 Years, 2008

Variable	Needed Care, AOR (95% CI)	Medications, AOR (95% CI)
Insured	0.32 (0.17, 0.62)	0.57 (0.25, 1.27)
Inconsistent coverage	5.99 (3.74, 9.58)	4.67 (2.71, 8.05)
Underinsurance
Prescriptions	0.93 (0.53, 1.65)	1.06 (0.54, 2.10)
Mental health	0.66 (0.47, 0.93)	0.77 (0.52, 1.16)
Dental	1.09 (0.77, 1.53)	0.69 (0.46, 1.02)
Type of insurance (Ref: employer-sponsored insurance only)
Medicare only	1.12 (0.58, 2.16)	1.13 (0.56, 2.28)
Medicaid only	0.77 (0.44, 1.34)	0.88 (0.48, 1.61)
Direct purchase only	1.86 (1.08, 3.21)	1.07 (0.53, 2.16)
Other/combination	0.56 (0.30, 1.04)	0.91 (0.49, 1.67)
Income
Household income	0.99 (0.990, 0.997)	1.00 (0.990, 1.002)
Income data missing	0.77 (0.46, 1.27)	0.75 (0.40, 1.40)
Usual source of care	1.23 (0.88, 1.71)	0.67 (0.42, 1.07)
Health status
Chronic physical condition	1.40 (1.00, 1.95)	2.64 (1.69, 4.13)
Chronic psychological condition	2.50 (1.40, 4.44)	2.38 (1.11, 5.10)
Both	0.71 (0.36, 1.38)	0.63 (0.27, 1.47)
Fair or poor health	1.25 (0.89, 1.76)	1.05 (0.71, 1.54)
Age, y (Ref: 18–29)
30–39	1.15 (0.69, 1.94)	2.06 (1.05, 4.03)
40–49	1.82 (1.11, 2.99)	1.70 (0.87, 3.30)
50–64	1.46 (0.89, 2.38)	1.85 (0.97, 3.52)
Women	1.74 (1.29, 2.33)	1.69 (1.18, 2.43)
Married	1.28 (0.96, 1.70)	0.72 (0.52, 1.00)
Hispanic	0.44 (0.30, 0.64)	0.71 (0.46, 1.11)
Medical debt
Problems paying bills	4.96 (3.34, 7.38)	6.16 (3.87, 9.81)
Current bills	3.04 (2.04, 4.52)	3.68 (2.31, 5.87)
Both	0.53 (0.29, 0.95)	0.43 (0.22, 0.85)
Nagelkerke R2	0.314	0.305

Note. CI = confidence interval.

Although income predicted both care and medication access problems when added in sequence (data not shown), after all explanatory variables had been entered higher income reduced the odds only of delayed care (AOR = 0.99). People reporting a chronic psychological or physical condition had increased odds of delayed care and medication use, but self-rated health status was not a significant factor. Women had higher odds than did men of both delayed care and delayed medications (AORs = 1.74 and 1.69, respectively). Self-identification as Hispanic or Latino was predictive of lower odds of delayed care (AOR = 0.44) but not of delayed medications.

Medical debt had a large effect on the odds of both delayed care and delayed medications even after control for all other variables. Among people who reported problems paying bills, the models predicted a 5-fold increase in the odds of delayed care (AOR = 4.96) and a 6-fold increase in the odds of delayed medications (AOR = 6.16). Among people who were currently paying off medical bills, the odds of delayed care tripled (AOR = 3.04), and the odds of delayed medications were even higher (AOR = 3.68). However, the odds of access problems among people who reported both that they had problems paying bills and that they had current medical bills were somewhat lower than the combination of the ratios would indicate, suggesting that these effects were not independent.

DISCUSSION

Our analyses contribute 2 major findings to the growing body of evidence on medical debt, insurance coverage, and access to care. First, simply having health insurance coverage does not appear to lower the odds of accruing medical debt. Second, medical debt is an independent and better predictor of delayed or missed medical care and medications than is insurance status.

Relative to residents of other states, Arizona residents appear to be somewhat better off in terms of medical debt and access to care. In 2008, 29% of Arizona residents reported problems with medical bills, debt, or both, as compared with 41% of nonelderly adults nationally in 2007.^2,3 In addition, 13% and 9% of the state's residents in 2008 reported delayed or missed care and prescribed medications, respectively, citing cost or lack of insurance as a reason. These rates are substantially lower than the 2007 national figure of 31% for each of these measures.^11,12

Although the relationships among medical debt, insurance coverage, and access to care have been documented, few studies have considered the relative contribution of each of these factors in the broader context of health status, age, income, and other characteristics. When examined alone, having health insurance coverage was associated with a lower prevalence of problems paying bills. However, after control for other variables simply having health insurance was no longer protective. Our data show that only 2 insurance-related factors affected the likelihood of medical debt.

First, gaps in coverage during the preceding year more than doubled the odds that an individual would have problems paying medical bills. The effect of inconsistent coverage could have been the result of medical bills incurred during the uninsured period, or it could be an indication of other financially related instability. Second, nonelderly adults who were covered solely through Medicare also had twice the odds of experiencing problems paying medical bills and having current medical bills. Because individuals younger than 65 years must be disabled or have end-stage renal disease to qualify for Medicare coverage, the impact on medical debt might be more a measure of the effect of health status than of insurance status.

Our analysis also confirms that insured status and medical debt are both independent predictors of delayed access to care, but it suggests that only medical debt predicts whether an individual will delay or forgo medications. Of note, although people who were uninsured were 3 times more likely to have delayed or not obtained needed medical care in the preceding year because of cost or lack of insurance, having current medical bills carried a similar independent risk, and having problems paying medical bills carried an even higher (5 times the odds) risk.

Of great concern is our confirmation of the impact of inconsistent coverage on both medical debt and access to care. Individuals who experienced coverage gaps were more than twice as likely as were those who did not to report problems paying medical bills, and they were 6 times as likely to report delayed care. Similarly, delays in obtaining and failure to obtain prescribed medication were most strongly predicted by medical debt and inconsistent coverage as opposed to insured status. Recent federal efforts to extend and subsidize COBRA (Consolidated Omnibus Budget Reconciliation Act of 1985) benefits along with federal and state efforts to expand coverage may help address this issue.^34,35

Finally, it is troubling to note that households with children were significantly more likely to report problems paying medical bills and to be currently paying off medical debt. It is also troubling that women and people with chronic physical or psychological conditions (arguably those at greatest need for health care services) were more likely to experience coverage- and cost-related problems accessing care.

By contrast, after control for other factors self-identification as Hispanic or Latino did not seem to affect the odds of medical debt or delayed medication use and actually lowered the odds of delayed care. The reason behind this finding is unclear. After controlling for income and insured status, culturally based social support^36,37 may make needed medical care seem more important and lower the tendency to delay or forgo this care. Also, this group may have had a lower perceived need for care.³⁸ In any case, this protective effect is worth further examination.

Limitations

Our models provide important preliminary insights into predictors of medical debt and access to care, but constraints inherent in the data set limit their explanatory power. Variables such as amount of nonmedical debt, availability of credit and savings, size of deductibles and copayments, and recent health care use might have added explanatory and predictive power to the medical debt models, and variables such as attitudes toward health care use may have added strength to the access models. Our analyses are also subject to the usual limitations of self-reported data. Despite these limits, our models were able to account for 12% to 31% of the variance in medical debt and access to care.

Conclusions and Implications

The implications of our results for health system reform are 2-fold. First, consistent coverage is essential, suggesting that health insurance should be made portable, universally available, or both. Second, although a primary purpose of health insurance is to provide protection from the financial impact of unexpected medical bills, once confounding factors are considered simply being insured does not appear to lower the odds of medical debt. Our analyses show that, at least in Arizona, health insurance is not protecting individuals and families from medical debt and that medical debt is a substantial and independent predictor of delayed or missed care and medication. Therefore, efforts to reduce large out-of-pocket costs of care are needed.

Arizona's medical debt and access problems appear to be somewhat lower than are the national estimates. However, given the challenges faced by many people as a result of the economic recession of the past few years^39,40 and new evidence about the financial impact of major illness,⁴¹ the results of our analysis may represent a troubling message for US health care in general.