Secondary Logo

Journal Logo


Update on the Seroepidemiology of Human Immunodeficiency Virus in the United States Household Population

NHANES III, 1988-1994

McQuillan, Geraldine M.; Khare, Meena; Karon, John M.; Schable, Charles A.; Vlahov, David

Author Information
Journal of Acquired Immune Deficiency Syndromes & Human Retrovirology: April 1, 1997 - Volume 14 - Issue 4 - p 355-357
  • Free


The third National Health and Nutrition Examination Survey (NHANES III) was the only U.S. Centers for Disease Control and Prevention (CDC) survey that provided an estimate of HIV infection in a representative sample of the U.S. household population. Estimates of HIV infection from NHANES III were used, along with data from other sources, to estimate the prevalence of HIV infection in the United States in 1992(1).

Data from the first 3 years of this 6-year national survey have been published previously(2). The prevalence of HIV from this half sample was 0.39%, with a population estimate of 547,000 infected individuals(95% CI 299,000-1,020,000). Nonresponse analysis was limited to demographic variables and demonstrated a lower response rate in young white men. In this article we present HIV prevalence estimates based on all 6 years of the survey and provide a more complete analysis of nonresponse to the survey than the previous publication. We also provide an estimate of the prevalence of HIV among cocaine users.

The results from this survey can be generalized to the civilian noninstitutionalized adult population 18-59 years of age. This household-based survey does not cover some groups who are at increased risk of HIV infection, including the homeless who do not reside in shelters, inmates in correctional facilities, and other institutionalized persons(3). The data presented here must therefore be interpreted carefully because the estimates are not corrected to include these populations.


Survey Design and Data Collection

The NHANES is a periodic national survey conducted by the National Center for Health Statistics. CDC. NHANES III was conducted from 1988 to 1994 in 81 randomly selected locations throughout the United States. The survey was divided into two 3-year surveys (phase 1 and phase 2) so that national estimates could be produced for each 3-year period as well as for the total 6 years. A total of 39,695 individuals ≥2 months of age were selected for the survey, with 31,311 (79%) agreeing to the examination component of the survey.

The NHANES III survey was based on a complex, multistage area probability sample design. Race/ethnicity was defined as non-Hispanic white, non-Hispanic black, and Mexican American. Individuals who did not self-select into these categories were called"other" and analyzed with the non-Hispanic white group. Children<5 years of age, persons ≥60 years of age, Mexican Americans, and black Americans were sampled at a higher rate than other persons. To produce national estimates, the results for each sample person were adjusted to account for the differential probabilities of selection and nonresponse(4,5). The HIV test was performed on serum collected from individuals ≥18 years of age. The HIV test and urine drug screening were performed on blinded specimens because of concerns about confidentiality.

Data analyses for this article were restricted to individuals 18-59 years of age because preliminary analysis of nonresponse demonstrated that individuals ≥60 years of age were less likely than younger persons to participate in the survey due to poor health. The possible bias introduced by health conditions unrelated to HIV/AIDS made estimation of HIV infection in this age group unreliable.

A urine specimen from individuals in the 18-59 year age range was tested for drugs of abuse during phase 2 of the survey (1991-1994). The urine drug result was linked to the HIV result along with age in 20-year age groups, race/ethnicity, and gender. Additional data were not linked to these results to maintain the anonymity of the participants.

Laboratory Methods

Serum specimens were tested for HIV-1 antibody using a U.S. Food and Drug Administration (FDA)-licensed enzyme immunoassay kit (Genetic Systems, Seattle, WA, or Organon-Teknika Corp., Durham, NC). Repeatedly reactive specimens were then tested by an FDA-licensed Western blot assay (Biotech/Dupont, Rockville, MD) and interpreted according to the ASTPHLD/CDC criteria(6). Urine was screened for cocaine using immunossay procedures (Abbott TDx, Carolinia, PR; Syva EMIT, San Jose, CA; and Roche RIA Systems, Belleville, NJ). All positive samples were confirmed using extraction/derivitization and gas chromatography-mass spectrometry. National Institute of Drug Abuse-established cutoff levels were used to indicate positive results.

Statistical Analysis

Prevalence estimates were weighted to represent the total United States population and to account for oversampling and nonresponse to the household interview and physical examination. The weights were further ratio adjusted by age, gender, and race/ethnicity to the United States population control estimates from the Current Population Survey adjusted for undercounts(4,5). Weights for phases 1 and 2 were adjusted to the March 1990 and March 1993 Current Population Survey estimates for the civilian noninstitutionalized population, respectively. Prevalence estimates for the total survey were based on combining one-half of phase 1 weights and one-half of phase 2 weights. Standard errors were calculated using SUDAAN(7), a family of statistical procedures for analysis of data from complex sample surveys.

A sensitivity analysis was performed using the method described by Brookmeyer and Gail to estimate the possible effect of nonresponse bias on the prevalence estimate(8). Male nonrespondents were assumed to have 2.5 times greater risk of HIV infection than male respondents. This ratio was based on information obtained in a follow-up survey of a household-based HIV survey conducted by the CDC in 1989(9). In that survey, men who initially refused to participate were two times more likely to report intravenous drug use and three times more likely to report male-to-male sex. The response rate for young or minority women was not markedly different from that of older or white women (Table 1). Therefore, nonresponse adjustments were not made to the estimate among women. A 95% confidence interval (CI) for the total adjusted estimate was computed under the assumption that the standard errors of the original and adjusted estimates were the same within the racial groups.

In phase 2 of the survey, 307 individuals had a urine specimen that was tested for drugs of abuse, but no matching blood specimen. Cocaine was chosen to assess the association between drug use and HIV infection and to impute HIV status in those who provided a urine sample but not a blood sample, because a larger number of participants had this drug in their urine (n = 205, six HIV positive) compared with morphine (n = 35, one HIV positive) and amphetamine (n = 9, zero HIV positive). In addition, two recent studies in San Francisco and Baltimore showed an association between cocaine use and HIV infection(10,11). Though the urine test cannot distinguish the route of cocaine administration, smoking of crack cocaine also has been associated with increased prevalence of HIV, especially in women(12). The HIV status among cocaine users and nonusers in the sample of 5,905 individuals from the second half of the survey was used to impute the HIV status among each of 307 persons who provided a urine specimen but not blood(13,14). Of these 307 persons, 19 were cocaine positive and 288 were cocaine negative. Individuals who used cocaine and did not have a blood sample were assumed to have the same proportion of HIV positive persons as the cocaine users with a blood sample within race/ethnic categories (Table 2). For the 288 nonusers of cocaine, the HIV sample proportions by race/ethnicity from the total survey were used in the imputation. A uniform random number generator was used to assign the HIV status. This process was repeated 100 times within the race/ethnicity, sex, and cocaine result categories. Final adjusted point estimates and CIs were computed from the means and standard errors obtained from the average of the 100 imputations using the augmented data set of 11,203 persons with a serum sample and the 307 who only had a urine sample.

Interview, examination, and phlebotomy response rates among persons 18-59 years of age by selected demographic characteristics(NHANES III, 1988-1994)
Prevalence of cocaine among NHANES III participants with a serum and urine specimen and HIV prevalence among cocaine positive persons


Prevalence Estimates

The HIV seroprevalence and population estimates based on the six years of NHANES III are presentedin Table 3. Of the 11,203 persons tested, 59 were positive for HIV antibody. This resulted in a weighted prevalence estimate of 0.32% (95% CI 0.20-0.51) and a population estimate of 461,000 (95% CI 290,000-733,000). Men were four times more likely to be infected then women (0.52%, 95% CI 0.32-0.85 vs. 0.13%, 95% CI 0.06-0.29). This male-to-female ratio differed by race/ethnicity, with black and Mexican-American men three times more likely to be infected than women, compared with white men, who were six times more likely to be HIV positive than white women.

Seroprevalence of HIV among adults living in households who are 18-59 years of age(NHANES III, 1988-1994)

Nonresponse Analysis

Table 1 presents the response rates for both the interview and examination components of the survey. For the 15,799 18- to 59-year-old participants selected for the survey, the overall response rate to the phlebotomy component was 71%. A lower phlebotomy response rate was observed for white/other men and the 40- to 59-year-old black men. The lowest response rate to the survey was observed in never-married white men 18-39 years of age (60%, data not shown). The potential effect of this nonresponse is summarizedTable 4. If the assumption that men who did not participate had a 2.5 times increased risk of infection was valid, the seroprevalence based on NHANES III data would increase to 0.45% with a population estimate of 651,000 (95% CI 409,000-1,000,000).

Adjusted seroprevalence of HIV in the total U.S. population assuming 2.5 times higher HIV infection among male nonrespondents by demographic characteristics

Among participants with both a serum and urine specimen available for HIV and urine drug testing, 1.7% were cocaine positive (Table 2). Men were significantly more likely to use cocaine than women, and black participants had the highest prevalence of cocaine use. The HIV prevalence among individuals who were positive for cocaine was 1.9%. HIV prevalence was higher among black women who used cocaine(6.5%) then among all black women (0.55%). No such difference was seen among other demographic groups. Cocaine use among participants without a blood specimen was twice as common as among those with a blood specimen (3.3% vs. 1.7%).

We used these data to impute the HIV status in the 307 individuals who had a urine result without an accompanying blood test. Eighty of the 100 random imputations added at most two HIV-positive persons who were predominantly black. The final combined estimate of HIV prevalence, an average of the 100 imputed estimates, was 0.27% (95% CI 0.15-0.49). The reduction in the estimate from the original estimate of 0.32% was due to the addition of white persons with high sample weights to the denominator and mainly black participants with low sample weights to the numerator of the proportion.


NHANES III is the only population-based national survey covering both men and women that provides an estimate of HIV prevalence. Although nonresponse to the survey may bias the estimate, NHANES routinely collects data needed to evaluate the potential for such biases and to estimate their magnitude. The estimates based on this survey also must be adjusted for populations not covered by the survey, but similar adjustments must be made for prevalence estimates based on other methods(1). The results from the survey clearly demonstrate a major epidemic among minorities in the United States.

The point estimate of HIV prevalence from the full NHANES III survey was 461,000 somewhat smaller than the point estimate of 547,000 obtained from the first half of the survey(2). However, the 95% CI of the estimate from the total survey includes the estimate from the first half of the survey. The number of infected persons found in the two phases of the survey were nearly equal (29 and 30, respectively;Table 3 ). The variability of the sampling weights between the two phases and the race/ethnicity of the infected persons account for the differing point estimates. The point estimate of 461,000 from the entire survey is also substantially less than the lower limit for the current plausible range for HIV prevalence in the United States of 650,000-900,000 persons(1). Evaluating the reason for this difference requires that we consider how the estimate based on NHANES III may have been affected by nonresponse to the survey and by the HIV prevalence in populations not covered by the survey.

There are two distinct reasons for nonresponse bias: (a) persons who declined to participate in the survey and (b) persons who were willing to participate but from whom a blood sample could not be obtained. The nonresponse adjustments to the sampling weights assume that respondents and nonrespondents within a weighting class have similar characteristics. The proportion who agreed to participate(as indicated by participation in the interview or the physical examination) was lowest among men (Table 1). We used results from another household survey to estimate that the potential bias in our prevalence estimate resulting from this nonresponse could be as large as 190,000 persons(9). Therefore, it is likely that the HIV prevalence estimate from NHANES III would have been substantially larger if all of those selected for the survey had agreed to participate.

Failure to obtain serum from some persons who participated in the survey could lead to a bias in the HIV prevalence estimate. This would be true if persons from whom serum could not be obtained were more likely to have behavioral characteristics associated with HIV infection than persons from whom serum could be obtained. During the survey, phlebotomists reported that at times a blood sample could not be obtained due to a difficult venipuncture. They suspected that there were occasions that this was due to intravenous drug abuse. Urine samples from participants in phase 2 of NHANES III were tested for metabolites of cocaine. The prevalence of HIV was higher only among black women who used cocaine (Table 2). Because blacks were oversampled in NHANES III, black participants have relatively small sample weights. As a result, failure to obtain serum samples from a relatively small proportion of black participants would have a relatively small effect on the prevalence estimate. Therefore, failure to obtain serum from some participants was not an important source of bias in the HIV prevalence estimate obtained from this survey.

The analysis of urine drug data from the survey provides evidence that black women who are cocaine users have a significantly higher risk of HIV infection. HIV prevalence was 6.5% for those who were positive for cocaine compared with 0.55% for all black women. This elevation was not seen in black men, and there were no HIV-positive results in the small number of individuals in the other racial groups who were cocaine positive.

Before comparing the HIV prevalence estimate from NHANES III with estimates based on other data, it is necessary to estimate HIV prevalence in populations outside the sampling frame for this analysis. The proportion of persons infected with HIV is relatively high among homeless persons(15) and among prisoners(16). In addition, the NHANES III sampling frame excluded institutionalized persons and residents of Puerto Rico and U.S. territories. Our analysis also excluded persons not 18-59 years of age. Based on data from a variety of sources, we estimate that there are ∼85,000-115,000 infected men and 25,000-50,000 infected women in these excluded populations(17). In addition, some severely ill HIV-infected persons may have declined to participate in NHANES III for personal health reasons. There were more than 100,000 HIV-infected persons who had already been diagnosed with AIDS opportunistic infections living in the United States in January 1993 (CDC, unpublished data). Refusal to participate by these persons also would be a substantial source of bias.

Two other methods have recently been used to estimate HIV prevalence in the United States(1). One method is back-calculation, which estimates the cumulative number of HIV infections and the probability distribution of infection times from known AIDS incidence and an estimate of the probability distribution for the time from HIV infection to AIDS diagnosis(17). The other method is based on estimates of the proportions of women infected in specified demographic groups obtained from a national seroprevalence survey of child-bearing women(18). These proportions are multiplied by the number of women (from U.S. Census Bureau data) to obtain estimates of the number of infected women of child-bearing age. AIDS surveillance data were used to extend these estimates to all women and to men(1). Both methods give similar ranges for the number of HIV-infected persons in the United States: ∽525,000-750,000 men and 120,000-160,000 women, or a total of 650,000-900,000 persons. The 95% CI for the number of HIV-infected women obtained from NHANES III (41,000-213,000, Table 3) contains the plausible ranges obtained from the other two methods. In contrast, the 95% CI for men obtained from NHANES III (225,000-600,000, Table 3) suggests a smaller number of HIV-infected men than do the ranges from the other two methods.

These comparisons suggest that there may be substantial nonresponse bias in the HIV prevalence estimate for men obtained from NHANES III. If we adjust for the potential effect of this bias, the 95% CIs for HIV prevalence from NHANES III are generally similar to, but wider than, the corresponding CIs from the other two methods for estimating seroprevalence. The wider CIs from NHANES III are a result of the relatively small number of HIV-infected persons found in this survey, a consequence of the fact that NHANES III did not cover some high-risk populations and was designed to estimate the prevalence of conditions substantially more common than HIV infection.


1. Karon JM, Rosenberg PS, McQuillan GM, Khare M, Gwinn M, Peterson LR. Prevalence of HIV infection in the United States. 1984-1992. JAMA 1996;276:126-31.
2. McQuillan GM, Khare M, Ezzati-Rice TM, Karon JM, Schable CA, Murphy RS. The seroepidemiology of human immunodeficiency virus in the United States household population: NHANES III, 1988-1991. J AIDS 1994;7:1195-201.
3. Centers for Disease Control. National HIV surveillance summary: results through 1990. Publication HIV/NCID/11-91/011. Atlanta, GA:U.S. Department of Health and Human Services, Public Health Service, 1991:2.
4. Ezzati T, Khare M. Nonresponse adjustment in a national health survey. 1992 Proceedings of the Survey Research Methods Section of the American Statistical Association, 1993:339-44.
5. Mohadjer L, Montaquila J, Waksberg J. National Health and Nurtition Examination Survey III: weighting and examination methodology. Prepared for National Center for Health Statistics, Hyattsville, MD. Westat Inc, February 1996.
6. Centers for Disease Control. Interpretation and use of the Western blot assay for serodiagnosis of human immunodeficiency virus type 1 infections. MMWR 1989;38(suppl 7):1-7.
7. Shah BV, Barnwell BG, Hurt PN, La Vange LM. SUDAAN users manual release 5.50. Research Triangle Park, NC: Research Triangle Institute, 1991.
8. Brookmeyer R, Gail MH. AIDS epidemiology: a quantitative approach. New York: Oxford University Press, 1994.
9. Centers for Disease Control. Pilot study of a household survey to determine HIV seroprevalence. MMWR 1991;40:1-5.
10. Chaisson RE, Bacchetti P, Osmond D, Brodie B, Sande MA, Moss AR. Cocaine use and HIV infection in intravenous drug users in San Francisco. JAMA 1989;261:561-5.
11. Anthony JC, Vlahov D, Nelson KE, Cohn S, Astemborski J, Solomon L. New evidence on intravenous cocaine use and the risk of infection with human immunodeficiency virus type 1. Am J Epidemiol 1991;134:1175-89.
12. Edlin BR, Irwin KL, Faruque S, et al. Intersecting epidemics—crack cocaine use and HIV infection among inner-city young adults. Multicenter Crack Cocaine and HIV Infection Study. N Engl J Med 1994;331:1422-7.
13. Kalton G, Kasprzyk D. The treatment of missing survey data. Surv Methodol 1986;12:1-16.
14. Rubin DB. Multiple imputation for nonresponse in surveys. New York: Wiley, 1987.
15. Allen DM, Lehman JS, Green TA, et al. HIV infection among homeless adults and runaway youth, United States, 1989-1992. AIDS 1994;8:1593-8.
16. Harlow CW. HIV in U.S. prisons and jails. Bureau of Justice Statistics, report number NCJ-143292. September 1993.
17. Karon JM. Methods for estimating HIV prevalence in the United States. Publication D821. Washington, DC: National AIDS Clearinghouse, U.S. Department of Health and Human Services, 1996.
18. Gwinn M, Pappaioanou M, George JR, et al. Prevalence of HIV infection in childbearing women in the United States: surveillance using newborn blood samples. JAMA 1991;265:1704-8.

Human immunodeficiency virus infection; National survey; Seroepidemiology; Prevalence; Urine drug testing

© Lippincott-Raven Publishers.