Trends in HIV Testing and Differences Between Planned and Actual Testing in the United States, 2000-2005

Background  Increasing the rates of human immunodeficiency virus (HIV) testing among groups not traditionally perceived as being at high risk has been advanced as a primary strategy in the effort to combat the HIV epidemic.

Methods  We conducted a pooled cross-sectional analysis of data from 146 868 participants aged 18 to 64 years in the 2000-2005 National Health Interview Surveys to describe longitudinal trends in HIV testing rates in the US population and differences between planned and actual testing across demographic and risk groups. Multivariable logistic models were estimated to assess correlates of perceived risk for HIV infection and planned and actual HIV testing. Difference-in-differences models examine how differences between planned and actual testing varied with demographic characteristics, perceived risk, alcohol consumption, depression, and health behaviors and access.

Results  Rates of HIV testing remained relatively unchanged from 2000 to 2005 (mean rates for lifetime and past year, 37% and 10%, respectively) and varied substantially by sex and race, with female and minority (nonwhite) populations more likely to get tested. Rates were higher in individuals reporting greater risks of HIV infection. However, even among respondents reporting medium or high risks of contracting HIV, less than 25% reported an HIV test in the previous year. Those with a higher perceived risk, more alcohol consumption, and more depressive symptoms had higher rates of both planned and actual testing but also demonstrated the greatest deficit of actual relative to planned testing.

Conclusions  In the United States, HIV testing rates remain low, nationally and in high-risk populations; low rates are likely contributing to a substantial number of undiagnosed cases of HIV. Despite above-average testing rates, populations considered to be at increased risk for HIV infection still demonstrate the need for improved access to and utilization of testing.

An estimated 1.1 million people in the United States are infected with the human immunodeficiency virus (HIV), and 24% to 27% are unaware of their infection.^1,2 Meanwhile, the HIV epidemic is increasingly transitioning out of traditionally high-risk groups, with increasing proportions of cases transmitted via heterosexual sexual intercourse and reported among women, racial and ethnic minorities (nonwhite persons), the poor, persons living in rural areas, and those living in the southern United States.^1,3-6 Initiatives to increase the rates of HIV testing, particularly among groups not traditionally perceived as being at high risk, have been advanced as a primary strategy in the effort to combat the HIV epidemic.^7,8 Indeed, the Centers for Disease Control and Prevention now recommends routine, opt-out testing for HIV (ie, an HIV test is performed unless the patient specifically declines) in all primary health care settings as well as annual HIV testing for persons at high risk of HIV infection; these recommendations are designed to decrease the number of infected persons unaware of their serostatus and therefore at risk of unknowingly infecting their contacts.⁸ Approximately one-third of US adults have ever been tested for HIV, and 10% to 15% have been tested in the past year.^9-11 Higher testing rates have been documented in groups considered to be at higher risk, such as men who have sex with men and injection drug users.¹²

In this study we used pooled data from 6 consecutive nationally representative cohorts of adults participating in the annual National Health Interview Survey (NHIS) to relate, at the individual level and in pooled cross sections, HIV testing rates to self-reported risk and to plans for HIV testing. In particular, we examined the difference between planned and actual testing and how this difference varied with demographic characteristics and with HIV risk and health behaviors.

The NHIS is an annual survey of the health of the civilian noninstitutionalized population of the United States, collected annually via in-person interview from 35 000 to 40 000 households.¹³ From 1 randomly selected (sample) adult in each household, the NHIS collects detailed information on a wide range of health-related topics.

This analysis draws on NHIS sample adult responses from survey years 2000 to 2005. Household response rates for these years ranged from 86.5% to 89.6%; response rates for the selected adult from each household ranged from 80.1% to 84.5%.¹⁴ All analyses were conducted on a pooled data set with appropriate adjustment to survey weights.¹⁵ Point estimates and standard errors account for the complex survey design.¹⁶ The pooled sample included 154 302 respondents aged 18 to 64 years, of whom 7434 respondents (4.8%) were excluded from multivariable analyses owing to missing values for dependent variables or key covariates.

This investigation was motivated by a conceptual model that posits that both planned and actual HIV testing behavior will be influenced by the following factors: (1) assessment of past HIV exposure, (2) expectation of present exposure risk (a combination of actual risk behaviors and subjective perceptions of risk), (3) general attitudes toward preventive health behaviors, and (4) barriers (access) to testing. Consistent with prior work,¹⁷ we postulated that a deficit of actual testing relative to planned testing reflects at least in part the presence of individual and structural barriers to testing.

To represent the first 2 domains of our conceptual model, we included as covariates both lifetime HIV risk and self-perceived current risk (defined in the next subsection, “Variable Definitions”). Also relevant to these 2 domains, we included measures of alcohol consumption and depressive symptoms. Alcohol use and depression have been consistently associated with increased HIV risk behaviors and also with differential risk perception^18-23 and thus should influence expectation of present exposure risk. As a measure of general attitudes toward preventive health behaviors as well as health care system access, we considered influenza vaccination.^24-27 Lifetime hepatitis B vaccination was considered to be both a marker for attitudes toward preventive behaviors as well as a marker of HIV risk, given the overlap between risk behaviors for HIV and hepatitis B infection. Finally, as a measure of a structural barrier to accessing HIV testing, we considered utilization of preventive health care services. Each variable is defined more precisely in the following subsection, “Variable Definitions.”

Participants were asked if they had ever received an HIV test, when the most recent test had been performed, and whether they were planning to get tested in the next 12 months. Those reporting a test were asked the reason for their last test. We categorized these reasons according to whether the respondents sought out the test (“voluntary”; eg, they wanted to find out if they were infected or had potentially been exposed to HIV), whether the test was part of routine medical care (“routine”; eg, part of routine medical checkup or prenatal care), or whether the test was mandatory (“mandatory” eg, for insurance or military service).

Participants were asked to indicate whether any of a list of lifetime HIV risk factors applied to them, without indicating which specific risk factor (receipt of clotting factors for hemophilia, a man who has had sex with men, injection drug use, trading sex for drugs or money, or having sex with someone who would affirm any of the previous items). Respondents who indicated that at least 1 of the statements was true were coded as having a lifetime HIV risk factor.

Respondents were also asked: “What are your chances of getting the AIDS virus? Would you say high, medium, low, or none?” Respondents who volunteered that they already had HIV were coded as “high risk” by NHIS interviewers; in the public-use data sets, these individuals cannot be distinguished from those who responded that they had a high risk.

Respondents who reported having had at least 12 drinks in their lifetime and at least 1 drink in the past year were classified as current drinkers. Current drinkers were classified as heavier drinkers if either (1) their alcohol consumption in the past year exceeded the current recommended maximum (a mean consumption of 14 drinks per week for men or 7 drinks per week for women²⁸) or (2) the respondent reported consumption of 5 or more drinks (binge drinking) on a mean of more than 2 occasions per month in the past year.

Respondents were asked the level (on a 5-point Likert scale) of 6 symptoms of depression in the past 30 days: feeling sad or blue, nervous, restless, hopeless, worthless, or that everything was an effort. The depression score, calculated as the mean of the 6 responses, ranged from 0 to 4 (coded so that a higher value indicated greater symptoms) and had a reliability coefficient of α = 0.87.

Two dichotomous variables were created indicating whether the respondent reported ever having been vaccinated for hepatitis B or having had an influenza vaccine in the past year. Respondents were classified as having a source of routine preventive care if they reported either receiving a routine medical checkup at a physician's office or a clinic in the past year or identified a physician's office or a clinic as the place they “usually” went to for preventive care. Individuals who reported usually going to a hospital or emergency department for preventive care were classified as not having a source of preventive care.

Other covariates included self-reported age, sex, race, ethnicity, educational achievement, marital status, and fair or poor health (vs excellent, very good, or good health), as well as census region of residence and the year of the survey.

We used logistic regression to model correlates of lifetime HIV risk factors and planned and actual testing, and we used ordered logistic regression to model correlates of self-perceived risk for contracting HIV. We examined the difference between planned HIV testing and actual HIV testing behavior, using a bivariate probit model to assess whether this difference varied across demographic and risk groups (analogous to the “difference-in-differences” method in the econometric literature^29,30). All models accounted for the complex NHIS survey design and the pooling of multiple survey years.^13,15,16 Analyses were performed using STATA statistical software (version 9.2; STATA Corp, College Station, Texas). The bivariate probit model was estimated with robust standard errors to account for error correlation across observations.

Rates of HIV testing remained low and relatively unchanged from 2000 to 2005 (Figure 1); modest increases in lifetime testing rates were observed in white and minority females. Minority females reported the highest rates of lifetime HIV tests. White males consistently reported the lowest rates of lifetime testing. The same pattern held for HIV tests in the past 12 months, with rates highest in minority females and lowest among white males.

Rates of lifetime and past-year testing were highly correlated with the presence of a lifetime HIV risk factor and with self-reported current risks of contracting HIV (Figure 2; P <.001 for all comparisons). Over 60% of respondents who reported a specific HIV risk factor or high current risk had ever been tested for HIV, and over 20% reported a test in the past year. Lifetime and past-year testing rates among respondents who reported “no risk” of contracting HIV were 36.2% and 9.1%, respectively. Plans for HIV testing during the next 12 months closely matched the pattern of reported testing during the past 12 months, ranging from 6.4% for those reporting “no risk” to 24.7% for respondents reporting “high risk” and 26.6% for those reporting a specific lifetime HIV risk factor (Table 1).

Of respondents who had ever been tested for HIV, 23.7% gave a reason for their last test suggesting they had voluntarily (intentionally) sought the test, for example, because of suspected exposure (Table 2), whereas 44.2% of tests occurred as part of routine medical care. Of note, more than 1 in 6 HIV tests were related to prenatal care (17%). An additional 20.9% of tests were required for insurance, marriage, immigration, or military service.

Self-reports of lifetime HIV risk factors were associated with younger age, being male, not being married, and fair or poor health, as well as with alcohol consumption, particularly heavier alcohol consumption, and with depression (Table 3). Hispanic individuals were less likely to report a lifetime HIV risk factor than white non-Hispanic individuals. Both flu shots and hepatitis B vaccinations were positively associated with the presence of lifetime HIV risk factors. Perceived current risk for HIV infection and plans for testing were generally associated with the same characteristics, but also with greater education and with race. Black non-Hispanic persons reported greater perceived risks of HIV than white non-Hispanic persons and were more likely to report plans to get tested. Hispanic persons and those of nonwhite racial groups reported lower perceived risks of HIV than white non-Hispanic persons but were more likely to report plans to get tested.

In multivariable models predicting lifetime and past-year HIV testing, all covariates were significant, suggesting large differentials in testing behavior across demographic characteristics and by alcohol consumption, depression, and vaccination behavior (see Table 4 for P values and confidence intervals). Rates of testing increased strongly with increased self-reported current risk of HIV infection and with the presence of 1 or more lifetime HIV risk factors. There was no differential effect of heavier alcohol consumption relative to light or moderate alcohol consumption on rates of either lifetime or recent testing. Consistent with expectations, vaccination history was associated with increased rates of testing for HIV, whereas lack of a source for preventive care was associated with decreased past-year testing.

When considering only voluntary tests, 2.1% of the sample reported an HIV test in the past year compared with 8.2% who reported an intention to get tested in the following year (Table 5), a difference of –6.1 percentage points between planned and actual testing. When considering both voluntary and routine tests, actual testing rates were only slightly below planned testing rates (7.4% vs 8.2%, a difference of –0.8%), and, when further including mandatory tests, actual testing rates exceeded planned testing rates (8.9% vs 8.2%, a difference of +0.7%). When considering only voluntary tests, the deficit of actual testing relative to planned testing was greater for those reporting a lifetime HIV risk factor (–16.0% vs –5.6% for those with no risk factor), those perceiving themselves to be at moderate or high risk of acquiring HIV (approximately –15.0% vs –5.0% for those perceiving themselves at no risk), those with greater alcohol consumption (heavier alcohol consumption, –8.8%; light or moderate consumption, –6.1%; no consumption, –4.8%), and those with depressive symptoms (–7.2% vs –4.4% for those with no depressive symptoms). Differences were lower for women relative to men and higher for minorities relative to white individuals.

These rank-orderings across subgroups remained consistent in direction and, for the most part, in magnitude as the measure of actual testing was expanded to include tests obtained during routine medical care, mandatory tests, and all tests (Table 5). Of note, those with no regular source of preventive care had a significantly greater deficit of actual relative to planned testing than those with such a source when considering voluntary and routine tests (–4.3% vs +0.2%; P = .02), but not when considering voluntary tests only (–8.2% vs –5.4%, P = .39).

Rates of past-year HIV testing remained constant and low throughout the study period. Large differences in testing rates according to race and sex remained relatively constant, with minority females reporting the highest rates of testing and white males reporting the lowest rates. Nearly half of HIV tests occurred as part of medical checkups or prenatal care, suggesting that policy initiatives to integrate testing into routine medical care have had some success.

Risk factors and behavioral correlates of risk were associated with greater rates of planned and actual testing. However, the difference between planned and actual testing was greater for those with greater HIV risk factors. Indeed, those with a lifetime HIV risk factor, with medium or high self-perceived HIV risk, and with heavier alcohol consumption were all less likely to actually get tested than to report an intention to get tested. The difference between planned and actual testing was greater for higher-risk than lower-risk groups regardless of whether we considered tests obtained for any reason or considered only voluntary tests.

This difference-in-differences analysis^29,30 suggests that individual and structural factors that inhibit the translation of testing intention into testing behavior may disproportionately affect groups at higher risk for HIV. Prior research has similarly examined differences between planned and actual testing and has described the individual- and structural-level factors that influence testing behavior, including concerns about confidentiality and discrimination, beliefs about test accuracy, social support, access to health care, and cost and convenience of testing.^17,31-33 The health beliefs model posits that health-related preventive behaviors result from the interaction of 4 dimensions: perceived susceptibility to the health threat, perceived severity of the health threat, perceived benefits of the preventive behavior, and perceived costs or barriers of the preventive behavior; later theorists have added self-efficacy as a fifth component.^34-36 In the present analysis, increased perceived susceptibility (risk) was associated with a decreased likelihood of intention translating into action. Alcohol use and depression, both of which affect perception of risks and benefits,^21-23 were similarly associated with decreased actual relative to planned testing, as was lack of access to preventive health care (a measure of a structural barrier). Flu vaccination was associated with higher HIV risk, perhaps indicating greater risk aversion or that individuals at higher risk of HIV may have higher utilization of health care services. Consistent with expectations, flu vaccination was associated with greater actual testing relative to planned testing.

Our conclusions are subject to a number of limitations common to survey-based analyses. The HIV risk measures in the NHIS may not capture all persons at increased risk of HIV (eg, heterosexual women with multiple sexual partners). In the question on perceived HIV risk, the high-risk category includes individuals who responded that they already have HIV. With an estimated 0.31% of the US population infected with HIV, of whom three-quarters are aware of their HIV status, affirmed HIV-positive individuals are likely to account for approximately one-third of those classified as high risk in the NHIS. Such individuals would be less likely to report a future intent to get tested, introducing a conservative bias in the planned-vs-actual difference for the highest-risk group.

In addition, recall bias and unwillingness to disclose sensitive information may downwardly bias reports of HIV testing and risk, respectively, whereas social desirability bias may artificially inflate reported testing intentions. Similarity between HIV risk and testing results from NHIS and other surveys using different methods supports the validity of the NHIS risk and testing questions.¹⁰ Furthermore, the difference-in-differences analytic approach used to examine between-group variation in the difference between planned and actual testing effectively cancels out biases that affect the compared groups equally. Although the absolute difference between planned and actual tests may not be directly interpretable as an “unmet need” for testing, the observation that this difference is greater for certain groups suggests that individual and structural barriers may operate differentially across HIV risk factors.

Of note, this repeated cross-sectional analysis compares planned testing in the year following the survey with actual testing in the year prior to the survey. Because we did not observe individuals at multiple points in time, we cannot compare an individual's expressed intention with that individual's subsequent action. Rather, we compare rates of planned and actual testing at the group level. The validity of such an analysis is vulnerable to external changes. For example, a coincident publicity campaign could cause next-year testing intentions to exceed past-year intentions (and actions). In the present analysis, we combined 6 consecutive annual cross sections and observed relatively stable testing rates over time. Furthermore, testing rates as well as differences between planned and actual testing remained stable both in analyses of repeated cross sections (at the population level) and in a multivariable choice model (at the individual level), suggesting that the difference in the reference time frame cannot account for the observed results.

Increased integration of HIV testing into routine medical care, as currently recommended by the Centers for Disease Control and Prevention, is likely to increase overall testing rates in the United States. Indeed, the present analysis indicates that nearly half of all HIV tests in 2000 to 2005 occurred as part of routine medical care. Although groups at higher risk of HIV (including those with heavier alcohol use and depressive symptoms) have higher rates of both planned and actual testing, it is precisely these groups who exhibit the greatest gaps between testing intention and action. These findings suggest that considerable potential exists to increase testing in higher-risk groups if individual and structural barriers can be identified and addressed. Alcohol and mental health treatment sites, for example, may represent a valuable opportunity to increase testing rates for higher-risk populations who exhibit a marked demand for testing. Although compelling arguments have motivated the current focus on general population testing,³⁷ such efforts should not come at the expense of ensuring access to and utilization of testing by higher-risk groups.

Correspondence: Brian Wells Pence, PhD, MPH, Center for Health Policy, Box 90253 Duke University, Durham, NC 27708 (bpence@aya.yale.edu).

Accepted for Publication: June 12, 2007.

Author Contributions: Dr Ostermann had full access to all of the study data and takes responsibility for the integrity of the data and the accuracy of the data analysis. Conception and design: Ostermann, Kumar, Pence, and Whetten. Acquisition of data: Ostermann, Kumar, and Pence. Analysis and interpretation of data: Ostermann, Kumar, and Pence. Drafting of the manuscript: Ostermann, Kumar, and Pence. Critical revision of the manuscript for important intellectual content: Ostermann, Kumar, Pence, and Whetten. Statistical analysis: Ostermann, Kumar, and Pence. Obtaining funding: Ostermann, Kumar, and Whetten. Administrative, technical, or material support: Pence. Supervision: Ostermann, Pence, and Whetten.

Financial Disclosure: None reported.

Funding/Support: This work was supported by grant No. 1R21 AA015052-01 from the National Institute of Alcoholism and Alcohol Abuse.