Are Patients at Veterans Affairs Medical Centers Sicker?:  A Comparative Analysis of Health Status and Medical Resource Use FREE

THE DEPARTMENT of Veterans Affairs (VA) is the largest single provider of health care in the United States. In 1997, the VA medical system treated 826,846 inpatients, provided 31.92 million outpatient visits, and had an operating budget of $17.55 billion.¹

Both the federal government and taxpayers have scrutinized the VA system in recent years. Critics claim that care delivered by the VA system is inefficient, expensive, and below community standards.^2- 3 These judgments are based on comparisons of VA medical care with non-VA care in terms of performance benchmarks developed by managed care organizations. Proponents of the VA claim that such comparisons are biased and invalid because of the sociodemographic and health differences between the VA patient population and the general patient population.

In today's environment of cost-efficient medical care, the VA has adopted managed care principles to stay competitive. The VA has also used managed care performance benchmarks for health care evaluation and resource allocation. However, a recent comparison of VA care with managed care showed that VA performance was poor when measured by these benchmarks.⁴ Peabody and Luck⁴ concluded that their findings could be biased by medical differences between the VA patient population and the general patient population. However, there is little data available to support this assertion.

In fact, a prior comparison of health status and health care use found no differences between veterans and the general patient population.⁵ A similar study found no differences in health status and medical resource use by female veterans compared with the general female patient population.⁶ Both studies concluded that VA patients are similar to the general patient population in health status and resource needs. Consequently, the VA administrators could extrapolate health data from the general population for health planning purposes within the VA. However, both studies were limited because they failed to differentiate between health status and resource use of those veterans who received medical care at the VA and those who received care in non-VA medical centers. The distinction between veterans who use VA services and those who use non-VA services is crucial in assessing health status and resource use of VA patients because only 10% of all veterans use VA medical services and the rest receive care at non-VA facilities.¹ In addition, the group of veterans who receive care at the VA is not a random sample of all veterans since entry into the VA is governed by specific eligibility criteria.

Prior to the eligibility reforms that went into effect in October 1998, entry into the VA was based on service-related medical conditions or disability or financial need. Therefore, a disproportionate number of all veterans who had service-related medical or psychiatric problems received medical care at the VA. In addition, those veterans who had nonservice-related medical conditions but had used up their financial resources purchasing medical care were also eligible to receive care at the VA. Based on these eligibility criteria, it seems likely that the VA patient population would have poorer health status and higher medical resource use than the general patient population. A recent analysis of data from the Veterans Health Study has supported some of these patient differences: veterans who received care at VA facilities were found to have poorer health-related quality of life scores compared with nonveteran patients.⁷ Results of this study were somewhat limited by the lack of a concurrent non-VA comparison group.

Thus, the primary objective of our study was to determine whether veterans who received care at the VA had (1) poorer health status and (2) higher medical resource use compared with the general patient population. A secondary objective was to determine whether veterans who received care in non-VA facilities had similar health status and resource use as the general patient population.

We conducted secondary data analyses of the "Access to Care" supplement of the National Health Interview Survey (NHIS) for the years 1993⁸ and 1994.⁹ The NHIS is a large national annual survey conducted by the National Center for Health Statistics. It is a principal source of information on the health of the resident, civilian, and noninstitutionalized population of the United States. The survey used for this study had a response rate of better than 85% for the years 1993 and 1994. With the use of appropriate sampling weights, the data from this survey is representative of the total US population.

The 1993 and 1994 NHIS surveys collected data on 177,466 respondents. We excluded respondents younger than 18 years, giving us a study population of 128,099 adult respondents. The NHIS survey identified respondents as being veterans, general population (ie, nonveterans), or unknown veteran status. All respondents were also asked to identify their "usual source of medical care" (physician's office, clinic at work, community clinic, county hospital clinic, private hospital clinic, emergency room, health maintenance organization clinic, VA clinic/hospital, military clinic, or other and unknown source of medical care). Based on responses to the questions on veteran status and the usual source for medical care, we categorized respondents into 4 groups. Respondents who identified themselves as veterans and identified a VA hospital or clinic as their usual source of medical care were labeled "VA veterans"; those who identified themselves as veterans, but reported a non-VA facility as their usual source for medical care were labeled "non-VA veterans"; respondents who identified themselves as nonveterans were labeled "general population"; and a fourth group, who was unsure of either their usual source of medical care or their veteran status was labeled "unknown."

Since the survey did not ask for any secondary sources of medical care, veterans who were dual users (used both VA and non-VA sources for medical care) could not be separately identified. It is estimated that 22% to 28% of veterans are dual users of VA and non-VA services.^10- 11 Studies have shown that dual users of VA care differ in socioeconomic status (SES) and medical care needs from veterans who receive all of their medical care at the VA. Dual users are more likely to be employed, have medical insurance, have a higher SES, and have greater health care needs.^11- 12

It is also known that veterans enrolled in Medicare and private insurance plans use VA benefits to receive medication and long-term care.¹³ Whether dual users have better health status (because of their higher SES%) or worse health status (because of greater health care needs) is not well documented.

The assessment of health status was based on self-reported health status and self-reported number of chronic medical conditions. Self-reported health status was measured on a Likert scale as excellent, very good, good, fair, and poor. This single measure of health status has been shown to be stable, reproducible, and highly correlated with morbidity^14- 18 and future mortality.^19- 20 The number of chronic medical conditions reported was used as a proxy measure of health status and a direct measure of comorbidity. The measure for chronic medical conditions has been objectively verified with physical laboratory examinations and medical records in other studies and is considered a valid measure of health status.^14,20 The 3 variables selected to measure medical resource use were (1) number of outpatient physician visits in the past year; (2) number of short-term care hospital admissions in the past year; and (3) number of days spent in the hospital for short-term care in the past year.

Control variables were selected based on their potential for being confounders or effect modifiers in our analyses of health status and resource use by veteran status. Control variables were entered into our multivariable model by using a manual stepwise approach with replacement.²¹ If a variable had a significant impact (either effect modification or confounding), then it was included in the final multivariable analyses. Based on these criteria, the following variables were included in our final model as control variables: (1) age in years (18-24, 25-44, 45-64, 65-74, ≥75); (2) sex; (3) race (white, black, and other); (4) educational status (elementary school, high school, high school graduate, college, college graduate, postcollege, and unknown); (5) annual family income (≤$19,999, $20,000-$34,999, $35,000-$49,999, and >$50,000); and (6) employment status in the last 2 weeks (currently employed, unemployed, and not in labor force).

The population sample for the NHIS was obtained through a sampling method involving both clustering and stratification. Because of this sampling design, direct application of standard statistical analysis would lead to underestimating the sampling variance.²² To address this issue we used the survey analysis module in STATA 5.0 (Release 5.0, 1997; Stata Corporation, College Station, Tex), which approximates sampling variances for survey estimates using the Taylor series linearization method and has been validated for analyses of complex survey data.²³ Descriptive statistics were generated to assess the relationship among veteran status, sociodemographic variables, health status, and resource use. Multivariable analysis using polytomous logistic regression models were used. We used a manual forward stepwise algorithm with replacement for model building,²¹ and hypothesis testing was conducted using the adjusted Wald test, for 2-sided hypotheses at a significance level of P<.05.

In our study sample of 128,099 respondents, approximately 14% of the sample (n = 18,338) identified themselves as veterans. Four percent (n = 666%) of the veterans identified a VA clinic or hospital as their usual source of medical care. The remaining veterans (n = 17,672) identified a non-VA facility as their usual source of medical care. Overall, 2% of the total study population (n = 3081) were unsure about their veteran status or their usual source of medical care, and were included in the unknown category. The unknown category was included in all analyses and showed no significant differences in health status and resource use from the general population. Hence, data on this group have not been presented but can be requested from the authors.

Both VA veterans and non-VA veterans were predominantly men and older than the general population (Table 1). Veterans Affairs veterans (VA veterans) were more likely to belong to a minority group compared with the general population. Non-VA veterans were more likely to be white compared with the general population. Veterans Affairs veterans had the lowest level of education, were more likely to be poor (49% reporting annual family income below $20,000), and were more likely to be unemployed than the general population. In comparison, non-VA veterans were similar to or slightly better than the general population in terms of education, employment status, and income.

With regard to health status, 19% of VA veterans reported that their health status was poor vs only 4% of the general population; 16% of VA veterans reported 5 or more medical conditions compared with 3% of the general population; and 42% of VA veterans reported 3 or more outpatient physician visits in a year compared with 20% of the general population. Veterans Affairs veterans also reported higher hospital admissions per year (21% reported 1 or more admissions) than the general population (7% reported 1 or more admission). Non-VA veterans were similar to the general patient population in terms of self-reported health status, number of medical conditions, and outpatient visits. Non-VA veterans had slightly higher hospital admissions than the general population (10% vs 7%, respectively, reported 1 or more admissions).

Veterans Affairs veterans were 14.7 times more likely to have poor health status than the general population, and 14 times more likely to have 5 or more medical conditions than the general population (Table 2). We found that effect modification due to age, sex, education, family income, and employment status was responsible for a large proportion of the increased risk for poorer health and higher number of medical conditions among VA veterans. However, even after controlling for these factors, VA veterans still had significantly poorer health status and more medical conditions than the general population (Table 2).

Non-VA veterans had small differences in health status and number of medical conditions when compared with the general population. However, these differences were mainly owing to the older age of non-VA veterans; in the adjusted analysis, non-VA veterans were similar to the general population in terms of health status and number of medical conditions (Table 2).

Since a larger percentage of the general population and non-VA veterans reported no medical care in the prior year compared with VA veterans (25% vs 14%), we compared medical resource use for only those respondents who were considered actively enrolled in the health system. This population consisted of respondents who reported at least 1 use of medical care in the preceding year (n = 96,777). This selection was done to avoid any potential for bias in analyses of medical resource use. Analyses of resource use indicated that VA veterans were 3.7 times more likely to have 3 or more outpatient physician visits in a year than the general population, and were also 5.4 times more likely to have 3 or more hospital admissions per year and 7.7 times more likely to have spent 21 or more days in the hospital (Table 3). After adjusting for confounding by age, sex, race, income, education, and employment status in the multivariable analyses of resource use, we found that differences in resource use between VA veterans and the general population were attenuated, but remained significant. Non-VA veterans had slightly higher resource use than the general patient population in the crude analyses, but these differences disappeared after controlling for age, sex, and SES differences (Table 3).

To determine if the higher medical resource use by VA veterans was due to preexisting poor health status of VA veterans vs inappropriate or inefficient care at the VA, we used a new multivariable model. In this model we controlled for self-reported health status and number of medical conditions, in addition to age, sex, race, income, education, and employment status. We found that after controlling for self-reported health status and number of medical conditions, inpatient medical resource use (number of hospital admissions and number of days in the hospital) was similar for VA veterans and the general population (Table 3). The difference in outpatient resource use was attenuated after controlling for health status and number of medical conditions, but remained statistically significant (odds ratio, 1.5).

In a large representative sample of the US population, we found that veterans who used the VA health system as their principal source of care reported poorer health and had a greater number of medical conditions than the general population. In addition, we found that medical resource use of VA veterans was higher than that of the general population of patients. This higher resource use was related to the age, SES, health, and comorbidity status of the VA veterans. After controlling for these differences, VA veterans and the general patient population had similar inpatient resource use. However, outpatient medical resource use remained significantly higher for VA veterans compared with the general population even after controlling for these modifiers. Non-VA veterans were similar to the general population in terms of health status and resource use, and therefore one can assume that non-VA veterans were less sick and used fewer resources than VA veterans.

The poorer health status of veterans was related to old age, low income, poor education, minority race both in our study and in others.^7,24- 26 Other factors such as health habits and combat-related physical and psychological conditions have also been shown to affect the health status of veterans. A study of Australian Vietnam veterans reported that increased combat exposure was significantly related to an increase in rates of acute and chronic health problems in veterans compared with age- and sex-matched members of the general population.²⁷ A study conducted on New Zealand Vietnam veterans also showed that veterans had diminished levels of psychological well-being, a greater number and severity of physical health symptoms, and lower self-reported health status than those in an age-matched general population.²⁸

The reason why not all veterans but only VA veterans are likely to belong to a lower SES and to have poor health is to be found in the eligibility criteria in place at the time of the 1993 and 1994 NHIS surveys. Since entry into the VA required a service-related medical condition or disability, a disproportionate number of all veterans who had service-connected medical or psychiatric problems received medical care at the VA. In addition, those veterans who had no financial resources or had used up their financial resources to pay medical bills became eligible to receive care at the VA based on financial need criteria. By these 2 mechanisms, veterans who belong to a lower SES (a predictor of poor health status) or have preexisting poor health were selected to receive care at the VA.

The use of health services by an individual is a function of predisposing, need, and enabling characteristics of that individual, as defined by the behavioral model of health services utilization.^29- 30 Predisposing characteristics (old age, lower income, lower education, and minority race) and need characteristics (poorer health, more medical conditions%) were strongly predictive of higher medical resource use by VA veterans in our study.

Study findings have implied that increased resource use by VA veterans is a function of enabling or organizational differences between VA and non-VA care. Comparison of resource use between a VA medical center and a large capitated multispecialty group practice reported that VA patients saw specialists 3 times more often, made 4 times as many emergency room visits, and had higher inpatient use and longer length of stay.⁴ An evaluation of short-term care admissions at a tertiary care VA medical center reported that only half of the admissions met health maintenance organization–developed appropriateness criteria.^31- 32 Another study that evaluated hospitalizations at 6 VA medical centers found that 26% of short-stay admissions were inappropriate.³³

Although these studies demonstrate the higher resource use at VA facilities than at non-VA facilities, their findings are potentially biased because they do not account for differences in sociodemographic and health characteristics between the VA patient population and non-VA patient population. In our study we found increased resource use by VA veterans as measured by outpatient physician visits and inpatient hospitalization rates and bed days of care. But after we controlled for differences in sociodemographic and health status, we found that medical resource use was similar for VA veterans and the general population. These results suggest that higher resource use by VA patients is not a reflection of organizational factors or inappropriate health practices. Instead, this higher resource use is an appropriate use of resources by a patient population that is older, has poorer health status, and more medical conditions than the general population of patients.

Our study used data already collected by the NHIS, so we were limited in our ability to identify dual user veterans (veterans who received care at both VA and non-VA facilities). The NHIS questionnaire did not collect data on dual or secondary sources of medical care. It is possible that dual users of VA care may have identified themselves as receiving care at the VA or non-VA facilities based on what they considered to be their "usual source of care." However, given the eligibility rules at the time of the survey, we suspect that dual users were more likely to receive their routine care from non-VA sources and use the VA for service-connected disability care only. Therefore, we believe that most dual users are included in the non-VA group of veterans. This misclassification would explain why only 4% of veterans in our study identified the VA as a source of care compared with the estimated 8% to 10% of all veterans who receive all or part of their medical care from the VA.^1,12 Comparisons of dual user veterans with veterans who receive all of their care at the VA have shown that dual users belong to a higher SES, have private medical insurance, and also have greater medical need.^11,13 Therefore, it is likely that our study overestimated the differences in SES and underestimated the differences in resource use between VA veterans and the general population.

Sampling and nonsampling errors were kept to a minimum by methods built into the NHIS procedures.³³ However, limitations in the NHIS sampling design led to the underrepresentation of the elderly population and the underestimation of hospitalizations for the elderly.³⁴ Another group that is not represented in the NHIS survey is the homeless population. Veterans Affairs patients make up a disproportionate percentage of the US homeless population, and homeless veterans have high health care needs.³⁵ Therefore, it is likely that veterans who use the VA are more likely to be underrepresented in the NHIS data because of their older age and higher rates of homelessness. This limitation could result in an underestimation of differences in health status and resource use of VA veterans compared with the general patient population in our study.

We found significant differences in health status and resource use between VA patients and the general population. During the period covered by this study, the VA patient population was older, less well educated, more frequently not in the work force, and poorer than the general population. However, even after controlling for these patient characteristics, we found that the VA patient population still had significantly worse self-reported health status and more chronic medical conditions than the general population.

Our study found higher medical resource use by the VA veterans, which is explained by sociodemographic characteristics and, more importantly, by the poorer overall health status and more chronic medical conditions of VA patients. Therefore, investigators and policy makers should not compare VA care with non-VA care based on resource use benchmarks such as number of visits or length of stay data, without accounting for differences between the VA patient population and the general patient population. Also, VA administrators and planners should not extrapolate health data from non-VA populations when estimating resource needs for VA patients. In particular, they must account for the adverse selection bias inherent in prior VA eligibility regulations that resulted in VA patients having both poorer sociodemographic characteristics and worse health status than either the general population or veterans who do not identify the VA as their regular source of care.

Accepted for publication June 7, 2000.

Presented in part at the national meeting of the Society of General Internal Medicine, Chicago, Ill, April 1998, and at the regional Midwest meeting of the Society of General Internal Medicine, Chicago, Ill, September 18-19.

Reprints: Zia Agha, MD, MS, Health Policy Institute, Center for Patient Care and Outcomes Research, Medical College of Wisconsin, 8701 Watertown Plank Rd, Milwaukee, WI 53226-0509 (e-mail: zia@mcw.edu).