Dr. Himelhoch, Dr. Medoff, and Dr. Dixon are affiliated with the Department of Psychiatry, University of Maryland School of Medicine, 737 West Lombard St., Baltimore, MD 21201 (e-mail: shimelho@psych.umaryland.edu). Dr. Dixon is also with the Veterans Affairs (VA) Capitol Health Care Network Mental Illness Research, Education, and Clinical Center, Baltimore. Dr. McCarthy, Ms. Ganoczy, and Dr. Blow are with the VA Serious Mental Illness Treatment and Evaluation Center, Ann Arbor, Michigan. Dr. McCarthy and Dr. Blow are also with the Department of Psychiatry, University of Michigan, Ann Arbor.

Individuals with serious mental illness are reported to be at high risk of HIV infection. Estimates from samples during the 1990s suggest that the prevalence of HIV among individuals with serious mental illness ranges from 3% to 23% (1,2,3,4,5,6,7,8,9,10). The wide variance in these estimates may be attributed to small samples, the use of convenience samples, differences in sampling frames, and inadequate adjustment for confounding effects of factors associated with HIV risk (11,12).

The only previous study that has attempted to determine an adjusted prevalence of HIV among persons with serious mental illness used Philadelphia Medicaid claims data from fiscal years 1994—1996 (13). The study found that compared with persons who did not have a mental disorder, those who had major affective disorders had 3.8 times the adjusted odds of having HIV and those with schizophrenia had 1.5 times the odds (13). Although this study controlled for age, race, ethnicity, and length of time on welfare, it did not adjust for substance use or marital status. Descriptive studies have indicated that individuals with serious mental illness may be at risk for HIV as a consequence of substance use disorder and of high-risk sexual behaviors (12,14,15,16). Previous studies have demonstrated that both substance use disorder and marital status are strong independent risk factors associated with HIV infection in the general population (17,18). Thus it is important to understand the influence of these factors on overall HIV risk among individuals with serious mental illness.

The goal of this study was to determine the recorded prevalence of HIV in a national population of patients with serious mental illness receiving care from Department of Veterans Affairs (VA) facilities and compare it with the prevalence in a national random sample of VA patients who did not have serious mental illness. This approach had two important strengths. First, having a large national sample of individuals with serious mental illness allowed us to overcome methodological limitations of previous studies—small samples and region-specific convenience samples—making it possible to generate more precise prevalence estimates. Second, having a comparison group of a random sample of VA patients who did not have diagnoses of serious mental illness allowed us to adjust for important HIV-related risk factors in order to more clearly understand the relationship between the diagnosis of serious mental illness and HIV.

Design

The study used a cross-sectional design with a national population of veterans with a diagnosis of serious mental illness and a national random sample of VA patients without serious mental illness who had at least three health care encounters at VA medical facilities during fiscal year 2002 (FY2002).

Data source

The VA National Psychosis Registry (NPR) is an ongoing registry of all VA patients for whom a serious mental illness diagnosis was recorded by VA providers at some point from FY1988 to the present. The NPR includes data from the VA's Patient Treatment File, census data files, and outpatient care files located at the Austin Automation Center in Texas. To be included in the NPR, patients must have a diagnosis of schizophrenia, bipolar disorder, or other psychoses (nonorganic). For patients with diagnoses of serious mental illness in a given fiscal year, patients are classified into three diagnosis categories. The categories are based on the following ICD-9 codes: schizophrenia (295.0—295.4, 295.6—295.9), bipolar disorder (296.0—296.1; 296.4—296.8), and other psychoses (297.0—297.3; 297.8—297.9; 298.0—298.4, 298.8—298.9).

Patients with substance-induced psychoses were not included in the definition of psychosis. If a patient received diagnoses within more than one diagnostic category of serious mental illness during FY2002, then an established algorithm was used to identify the primary serious mental illness diagnostic category (19). FY2002 was chosen as the baseline year of a larger longitudinal study that builds on the Serious Mental Illness Treatment, Research, and Evaluation Center's comprehensive registry data, which include a random sample of all VA patients in that year. The exposure period of the study was all of FY2002.

The NPR includes comprehensive data on utilization of VA health services, including diagnoses, medication use, and costs as well as information regarding patient and treatment facility characteristics. This data set has been previously used to report on mental health epidemiology, patterns of mental health services utilization, and use of psychotropic medications (20,21).

A sample of 100,000 veterans included in the Serious Mental Illness Treatment Research and Evaluation Center's national random sample of users of the VA health care system was used to derive a comparison group. From this sample we identified patients who met inclusion criteria and did not have a serious mental illness diagnosis in FY2002. These individuals constituted a comparison group without serious mental illness. Inclusion in either the group with serious mental illness or the group without serious mental illness was limited to individuals with at least three VA health encounters in FY2002. We limited our sample to veterans with three encounters because individuals with fewer VA encounters may have had fewer opportunities to receive an HIV test or to have an HIV diagnosis recorded. With this criterion, our final sample included 191,625 individuals with diagnoses of serious mental illness and 67,695 individuals without such diagnoses in FY2002.

Decedents were included in both the NPR data as well as the random sample of patients without serious mental illness. Among study patients, 2.9% of those with serious mental illness and 2.0% of those without serious mental illness died before the end of FY2002.

Study patients with serious mental illness had a mean±SD of 25.0±33.9 service encounters in FY2002 (median=14), and those without serious mental illness had 8.7±12.9 encounters (median=5).

Definitions

We used ICD-9 codes from inpatient and outpatient administrative encounter data to identify patients who had a diagnosis of HIV infection during FY2002. ICD-9 codes used to identify HIV infection included 042.0—042.9, 043.0—043.9, 044.0—044.9, 795.8, and V08 (22). HIV diagnoses were identified by examining g all VA inpatient and outpatient utilization records for FY2002. Thus we use the phrases "recorded prevalence of HIV" and "risks of recorded diagnoses of HIV."

Patients' measures included age, gender, race, marital status, region, locale, military service-connected disability, and homelessness in FY2002. Race-ethnicity was categorized into four groups: black, Hispanic, and white and a single category for other, unknown, or missing. The U.S. region where the veteran received VA services was classified into four mutually exclusive areas: Central, Northeast, South, and West (23). Locale was defined by metropolitan statistical areas and reflected whether a patient resided in an urban or rural setting. Homelessness was based on diagnoses for which the code V60 was used or on confirmed use of specific VA services for homeless veterans (24).

Finally, a substance use disorder was defined in terms of ICD-9 codes for abuse of or dependence on alcohol, cocaine, opiate, cannabis, and combinations of drugs (303.0, 303.9, 304.0—9, 305.0, and 305.2—9). The substance use disorder variable was based on administrative data of patients who had substance abuse or dependence diagnoses recorded in their encounters. The potential for measurement error in assessing the presence or absence of conditions by using administrative data exists; however, VA diagnosis data has been examined in several studies and found to be in close agreement with chart data (25,26,27).

Analysis

Univariate distributions included percentages for dichotomous and categorical variables and means for continuous variables. Comparisons of means were made by using the Wilcoxon rank-sum test, and comparisons of percentages were made with chi square tests.

Bivariate and multivariate logistic regression analyses were used to examine the relationship between having a diagnosis of serious mental illness and the outcome of interest—diagnosis of HIV. These models were used to estimate relative odds ratios and 95% confidence intervals and to adjust for differences in demographic risk factors and risk factors related to HIV. The serious mental illness variable was analyzed as a dummy variable comparing risk among patients with diagnoses of schizophrenia, bipolar disorder, or other psychoses and those without a diagnosis of serious mental illness. When no significant differences were found by using the dummy variable, the composite binary variable comparing those with serious mental illness diagnoses with those without such diagnoses was used. Because HIV is a rare outcome, odds ratios are likely to closely approximate relative risks (28).

Because previous studies have demonstrated that both substance use and marital status are strong independent risk factors associated with HIV infection, we used these variables as surrogates to adjust for the effects that injection drug use and high-risk sexual behavior have with respect to HIV infection (17,18). Adjusted probabilities representing the adjusted prevalence of HIV were computed by back-transformation of the adjusted odds ratios by using the following characteristics: male, 50 years of age, white, married, not homeless, South region, urban locale, and service-connected disability.

We evaluated interactions between serious mental illness diagnoses and substance use disorder and between serious mental illness diagnoses and marital status to investigate whether subgroups of individuals with serious mental illness may be differentially at risk of HIV. Statistical analyses were completed using SAS for Windows, release 8.02. All reported p values are two-sided.

Demographic and clinical characteristics

Of the 191,625 VA patients with a serious mental illness diagnosis in FY2002, 46.5% (N=89,189) had schizophrenia as their primary serious mental illness diagnosis, 34.4% (N=65,983) had a diagnosis of bipolar affective disorder, and 19.0% (N=36,453) had a diagnosis of other psychotic disorders (Table 1). Compared with those without diagnoses of serious mental illness (N=67,965), those with serious mental illness diagnoses were significantly more likely to be younger, female, black, and Hispanic, to have a co-occurring substance use diagnosis, to live in an urban area, and to be homeless. Those with serious mental illness diagnoses were also less likely to be married.

Crude recorded prevalence of HIV

The crude recorded prevalence of HIV among those with diagnoses of serious mental illness was 1.0%. No significant differences were found in the recorded prevalence of HIV between the serious mental illness subgroups (1.0% for all subgroups). Among patients without diagnoses of serious mental illness, the crude recorded prevalence of HIV was .5%. Compared with patients without diagnoses of serious mental illness, those with such diagnoses had nearly twice the odds of having a recorded diagnosis of HIV (OR=1.94, CI=1.73—2.17).

As shown in Table 2, the crude odds of having a recorded HIV diagnosis for patients with schizophrenia, bipolar disorder, or other psychoses were each nearly twice as great as those for patients without serious mental illness diagnoses (schizophrenia, OR=1.90; bipolar disorder, OR=2.00; and other psychoses, OR=1.92). The crude odds of having a recorded HIV diagnosis for patients with substance use diagnoses was over three-and-a-half times as great (OR=3.69) as those for patients without substance use diagnoses. The crude odds of having a recorded diagnosis of HIV for patients who were never married or divorced were two to six times as great as for those who were married (never married, OR=5.88; divorced, OR=3.58; and widowed, OR=2.10).

Table 2 also shows that the crude odds of having a recorded diagnosis of HIV was twice as great among men as among women (OR=2.12). Compared with white patients, the crude odds of having a recorded HIV diagnosis was three to nearly five times as great for Hispanic and black patients (Hispanic, OR=3.25; black, OR=4.90. Patients who received care in urban settings had three and a half times the odds of having a recorded diagnosis of HIV (OR=3.49) compared with patients in rural settings, and those who were homeless had more than three times the odds of having a recorded diagnosis of HIV (OR=3.28) compared with those who were housed.

Adjusted recorded prevalence of HIV

After the analyses adjusted for age, race, gender, marital status, substance use disorder, disability status, homelessness, region, and locale, patients with bipolar disorder diagnoses were no more likely than those without diagnoses of serious mental illness to have a recorded diagnosis of HIV (OR=1.08), and patients with other psychoses were 18% more likely than those without serious mental illness diagnoses to have a recorded diagnosis of HIV (OR=1.18).

We observed a significant interaction between schizophrenia diagnoses and substance use diagnoses. Compared with patients who had neither schizophrenia nor a substance use diagnosis, those with schizophrenia without a substance use diagnosis were significantly less likely to have a recorded HIV diagnosis (OR=.49) and those with a substance use diagnosis were significantly more likely to have a recorded HIV diagnosis (OR=1.22) (Table 2). The adjusted recorded prevalence of HIV stratified by serious mental illness diagnosis is presented in Figure 1.

Interactions between a substance use disorder diagnosis and bipolar disorder or other psychosis were not significant. Also, interactions between schizophrenia, bipolar disorder, or other psychosis and marital status were not significant.

Table 2 also shows that the adjusted odds of having a recorded diagnosis of HIV for patients with substance use diagnoses was nearly one-and-a-half times as great as those for patients without a substance use diagnosis (OR=1.42). Patients who were never married, divorced, or widowed were two to nearly four times as likely as those who were married to have a recorded diagnosis of HIV (never married, OR=3.83; divorced, OR=2.50; or widowed, OR=2.14).

Male patients were two to three times more likely than female patients to have a recorded HIV diagnosis (OR=2.74). The adjusted odds of having a recorded diagnosis of HIV for black or Hispanic patients were more than three times higher than those for white patients (black, OR=3.45; Hispanic, OR=3.30). Patients who received care in urban settings had twice the odds of those in rural settings of having a recorded diagnosis of HIV (OR=2.12). Finally, those who were homeless were as likely as those who were housed to have a recorded HIV diagnosis (OR=1.07) (Table 2).

Among a national population of VA patients, we found that the crude odds of having an HIV diagnosis were twice as high among patients who had diagnoses of serious mental illness as among patients who did not have serious mental illness diagnoses. This relationship was observed overall and by diagnostic subgroup of serious mental illness. This finding is consistent with results of many previous studies that have found elevated rates of HIV among individuals with serious mental illness. Although the recorded prevalence of HIV among those with serious mental illness in our study was lower than that in previous studies, an important strength of our findings is that they are based on a national sample of individuals with serious mental illness. As a result, our findings may not be subject to the biases associated with convenience samples or differences in sampling frames. Our results are also consistent with results of a previous study that used administrative data; this study found that the crude period prevalence of HIV was approximately twice as high among Medicaid beneficiaries with schizophrenia as among beneficiaries without serious mental illness (13).

In addition, we found that adjusting for HIV risk factors and demographic characteristics considerably reduced—and even inverted—the association between diagnostic subgroups of serious mental illness and HIV. Specifically, we found that after the analyses adjusted for HIV risk factors and demographic characteristics, patients with bipolar disorder were as likely to have a diagnosis of HIV as those without serious mental illness and that those with other psychoses were 18% more likely. This considerable reduction in the association between serious mental illness diagnoses and HIV highlights the importance of adjusting for risk factors associated with HIV. This is most clearly the case for individuals with schizophrenia, among whom we found a significant interaction between substance use disorder and schizophrenia diagnosis. Specifically, compared with patients who did not have serious mental illness, those with co-occurring schizophrenia and a substance use disorder were 22% more likely to have HIV and those with schizophrenia who did not have a co-occurring substance use disorder were 50% less likely to have HIV. In other words, not having a substance use disorder appears to be protective among persons with a diagnosis of schizophrenia.

Previous studies suggest that people with schizophrenia tend to have lower overall rates of sexual activity than the general population (12), which may highlight the impact that substance use has on the risk of contracting HIV and on engaging in high-risk sexual behavior. Alternatively, some studies suggest that among persons with schizophrenia, those with substance use disorders have fewer negative symptoms than those without substance use disorders (29,30). Thus it is possible that negative symptoms, such as apathy, decreased sociability, and poverty of speech, may in fact be the moderating factor accounting for our finding that patients with schizophrenia who did not have a substance use disorder were significantly less likely to be infected with HIV.

Alternatively, our finding that patients with schizophrenia without a co-occurring substance use disorder were less likely to have an HIV diagnosis may be the result of underdiagnosis or undercoding of HIV diagnoses. Although worsening psychotic symptoms may be a barrier to receiving HIV testing, a study by Desai and Rosenheck (31) found that among a cohort of homeless individuals with serious mental illness who were enrolled in a case management program, the likelihood of being tested for HIV was independently associated with more severe psychiatric symptoms. Finally, because persons with schizophrenia and a co-occurring substance use disorder are more likely to be at greater risk of not using medical services, we believe we probably underestimated, rather than overestimated, the effect of the interaction—that is, patients with schizophrenia and a co-occurring substance use disorder are likely to be at even greater risk for HIV infection.

It is particularly noteworthy that the relative odds of HIV among individuals with a diagnosis of schizophrenia and a co-occurring substance use disorder were substantially lower in the adjusted model, which is consistent with the lower rates we found in the adjusted model for those with bipolar disorder and for those with other psychoses.

We also found that risk factors associated with HIV in our study were consistent with those that have been reported for the general population (32) and for VA samples (33). First, the prevalence of HIV in the control group without serious mental illness in our study is consistent with the prevalence of HIV of .5% reported by the VA Immunology Case Registry in FY2002 (33). Our sample was similar to the VA Immunology Case Registry, in that it was predominantly male and older than the general non-VA population.

Second, we found that being male, black or Hispanic, and unmarried, having a diagnosis of a substance use disorder, and receiving services in an urban setting were associated with a greater risk of HIV. Although homelessness was positively associated in the crude analysis with having an HIV diagnosis, it was no longer significant after the analysis adjusted for a diagnosis of serious mental illness and a co-occurring substance use disorder. This may reflect the positive associations between being homeless and having a severe mental illness (34). HIV has been noted to be particularly endemic in urban locales in the coastal regions of the United States, and thus it is not surprising that those who received services in the Central region were significantly less likely to have HIV.

The study had some limitations. First, because it is cross-sectional, we cannot make strong claims about the direction of causation. Although HIV may cause people to have depressive and psychotic symptoms (35), research suggests that in most cases, psychotic disorders precede HIV infection (36). Second, the study was limited by factors associated with using administrative data sets. Because substance use disorders are frequently underdiagnosed, it is possible that we underestimated the frequency of these disorders in the sample. It is also possible that we underestimated the prevalence of HIV, because diagnosis data in administrative data sets may not fully identify individuals with HIV who are not actively engaged in HIV treatment or who may choose to receive that treatment from a non-VA provider.

We attempted to decrease the likelihood of this limitation by restricting our analysis to individuals with at least three health care visits during FY2002. Compared with those with three or more visits, those with two or fewer visits were significantly more likely to be female, to have no race recorded, and to be married; they were less likely to have an indication of homelessness, a substance use disorder, or a service-connected disability. By limiting the analysis of HIV prevalence to patients with at least three VA encounters in FY2002, we excluded patients who may have received most of their care (and illness diagnoses) from non-VA providers. Patients with fewer than three encounters or with none at all had fewer or no opportunities to have had HIV diagnoses recorded in the VA data in FY2002. Finally, we were unable to distinguish receipt of an HIV diagnosis from receipt of an HIV diagnosis as a part of HIV treatment management in this sample (Bowman C, HIV Query Group, personal communication, May 2006).

Third, because the results of this study are based on data from veterans who were receiving care in VA facilities, it is unclear to what extent the results may be generalized to veterans who are infrequent users of VA services or to the nonveteran population. Rosenheck and colleagues (37), using data from the Schizophrenia Patient Outcomes Research Team (PORT) and data from the VA extension of the PORT, compared the treatment of schizophrenia in VA and non-VA clinics and found that persons with schizophrenia treated in VA clinics were more likely to be male and older and to have higher incomes but were no different in race, marital status, or education than persons with schizophrenia treated in non-VA clinical settings. They also reported no differences with respect to alcohol or substance use, incarceration, symptom distress, satisfaction with providers, and community adjustment compared with individuals with schizophrenia who were receiving care in non-VA clinics. They did note that those receiving care in the VA system were less likely to use community-based psychosocial treatment and more likely to rely on hospital treatment than non-VA patients. Young and colleagues (38) compared a random sample of outpatients with schizophrenia at a VA facility in California with a random sample of outpatients with schizophrenia at a regional mental health clinic in California. They found that VA patients were more likely to be older and male and to have higher incomes.

Although we did not have information on age of onset of symptoms, it appears that patients with schizophrenia in treatment may not differ in social functioning compared with those receiving non-VA services (37). Although the VA may be the provider of last resort for veterans without serious mental illness, it is unclear whether this is the case for individuals with serious mental illness (37). Finally, because some patients may use two systems of care—that is, Medicare and VA services—it is possible that our results are biased because we were unable to capture medical service use in other systems of care. Further research is necessary to clarify this issue.

Fourth, we did not have specific variables to adjust for the mediating roles of injection drug use and high-risk sexual behavior, which are known to be causally linked with serious mental illness and transmission of HIV. Instead, we used the more distal variables of substance use disorder and marital status to account for these risk factors. We used these variables, as have previous studies (17,18), because it is difficult to obtain reliable estimates of stigmatized behavior, such as high-risk sexual behavior. However, our analysis indicates that these proxy measures were strongly associated with HIV diagnosis, and therefore we believe they served as reasonable alternatives.

Fifth, if the observed differences in age between the group with serious mental illness and the control group were related to differential early mortality from HIV, we may have underestimated the prevalence of HIV among patients with serious mental illness. However, sensitivity analyses showed that the serious mental illness group in fact had a higher prevalence of HIV in the older age groups than did the control group without serious mental illness. If differential mortality from HIV were associated with the observed age difference, the trend would be the reverse, with higher prevalence among those with serious mental illness at younger ages.

Sixth, although the absolute difference in the adjusted probabilities of having a recorded HIV diagnosis is small, on a population level the difference would represent a considerable reduction and would likely have considerable clinical merit. Seventh, information about the frequency of HIV testing in the VA system of care is currently not published. However, the VA's policy (39) is to "urge all VA providers to make risk assessment and testing for HIV, sexually transmitted diseases, and other chronic viral infections a routine part of medical care." Finally, several reports suggest that there is acceptable concordance between VA administrative data and medical record data (25,26,27).

Despite the elevated crude prevalence of HIV, multivariate analyses suggested that HIV-related risk factors may underlie the observed associations between HIV and each of the serious mental illness diagnoses. For patients with schizophrenia, this study is the first to demonstrate significantly lower HIV risk in the absence of substance use disorder.

Given the continued risk for HIV in this vulnerable population (40), ongoing efforts by mental health providers to provide consistent primary and secondary HIV prevention education is warranted (41). This may be particularly important for individuals with schizophrenia who are known to be abusing drugs or alcohol, because the findings indicate that patients with schizophrenia who also had a substance use disorder were at greater risk of being diagnosed as having HIV. It also may be particularly important for mental health professionals to monitor (42) and maximize psychiatric treatment interventions that may reduce the risk during an acute psychiatric episode of high-risk sexual behavior or use of illicit substances—for example, optimizing treatment of bipolar disorder to prevent mania. Future research is needed to examine how serious mental illness is linked to HIV risk behaviors.

The Centers for Disease Control and Prevention recommends making HIV testing a routine part of medical care in clinical settings that serve individuals with a high prevalence of HIV (43). Because the prevalence of HIV among VA patients with serious mental illness is 1%, interventions aimed at screening and reducing the risk of blood-borne infectious diseases may be particularly important. Results from an ongoing randomized, controlled trial called STIRR (Screening and Testing for HIV, Hepatitis C and Hepatitis B, Immunization for Hepatitis A and B, and Risk Reduction Counseling) that specifically targets individuals with serious mental illness in outpatient mental health clinics may be an important and innovative first step to ensuring that HIV testing is conducted in this vulnerable population (44).

The study was funded by VA Capitol Health Care Network Mental Illness Research, Education, and Clinical Center and by grant 1K-23-DA019820-01 from the National Institute of Drug Abuse to Dr. Himelhoch.

The authors report no competing interests.