Our definition of heterosexual was created to best correspond to the HIV transmission category used for surveillance [1] as our ultimate purpose for this analysis was to calculate disease rates. “Transmission category” is the HIV surveillance term for the classification of cases among those aged 13 years or older that summarizes a person’s possible HIV risk factors; the summary classification results from selecting, from the presumed hierarchical order of probability, the single risk factor most likely to have been responsible for acquiring HIV infection. Persons with > 1 reported risk factor are classified in the category listed first in the hierarchy. The exception is men who had sexual contact with other men and injected drugs; this group makes up a separate transmission category. Persons whose transmission category is classified as male-to-male sexual contact include men who have ever had sexual contact with other men, including men who have ever had sexual contact with men and with women. Persons whose transmission category is classified as heterosexual contact are persons who have ever had heterosexual contact with a person known to have, or to be at high risk for, HIV infection (e.g., a person who injects drugs).

Following the hierarchy used for transmission category, we calculated the proportion heterosexual with no other HIV risk behaviors from survey data (described below) by excluding the following: 1) those who reported never having had sex, 2) males who reported sex with another male, 3) males and females who ever injected drugs, and 4) females who only reported sex with female partners (i.e., never had sex with a man). The remaining proportion, excluding those with missing data, was considered heterosexual (Fig 1). Our choice to use lifetime (ever) behavior to define heterosexuals corresponded to the transmission category definition, which is behavior since 1977 [1].

The three national population-based surveys included were the National Survey of Family Growth (NSFG, 2006–2010), the National Health and Nutrition Examination Survey (NHANES, 2009–2010), and the General Social Survey (GSS, 2010). These surveys and general question wording are described in Table 1; website addresses are provided for further information about the sampling methods, human subjects review, specific question wording, response rates, and weighting.

For NSFG and GSS, we determined the proportion of heterosexuals and the variance (standard error) using public use datasets. We obtained permissions to use restricted data for NHANES to include persons 15 and older as the public use dataset only includes persons aged 18 years and older. We used 15 years as the lower age limit for NHANES to match the lower age limit of NSFG. For GSS, we used 69 years as the upper age limit to match NHANES; the lower age limit for eligibility in GSS is 18 years. Because NSFG data on lifetime injection drug use were not available in the public use dataset, we used population proportions of persons who inject drugs from previous analyses [3] to adjust the estimated population proportion heterosexual for NSFG.

For each survey, we calculated the proportions of heterosexuals for the overall population. Stratified analyses were conducted by sex, race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, and all others), and age group (15–24 years, 25–34 years, 35–44 years, 45–54 years, and 55–69 years) and for males and females by race/ethnicity and age group. For these analyses, we used SUDAAN software version 9.1 (RTI International, Research Triangle Park, NC) [5] to account for the complex sample designs. Differences in stratum-specific estimates were determined by non-overlapping confidence intervals.

To combine the 3 distinct estimates into a combined measure, we applied a meta-analytic method that has recently been extended to survey data [6]. We first multiplied each survey estimate by a weight inversely proportionate to its variance, summed the weighted estimates across studies, and then divided by the sum of the weights.

The studies included for the meta-analysis were sufficiently homogeneous in terms of sampling methods, participants, and outcomes to provide a meaningful combined measure. All were national probability surveys designed to make inference to the U.S. household-based population, and collected self-reported data on sexual behavior. Despite these similarities, it is possible that differences in characteristics of the surveys, such as question wording, could result in heterogeneity. We selected random effects models for our analyses because the models assume the studies are a random sample [7], a type of inference that fits the purpose of our study which is to estimate the population proportion of heterosexuals. In our analysis, the estimates of the proportion heterosexual are not identical across surveys but rather have a distribution (under the random effects model assumption); the combined estimate describes the average of the survey estimates and the confidence interval provides an indication of the spread of the distribution of population proportion estimates of heterosexuals. The meta-analysis method developed by Rao et al [6] adds a between-studies variance term in deriving an overall estimate. Heterogeneity of estimates across surveys is indicated with the Q statistic [6] and Higgins' I² index [8]. The Q statistic follows a chi-square distribution and assesses whether observed differences in results are compatible with chance alone. I² describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error [9]. Values of the Q statistic indicated that the between-studies variance term was a statistically significant source of variability. Stratified analyses by sex and age allowed us to further address sources of heterogeneity across surveys.

We carried out all estimates per Rao’s method using Microsoft Excel (2007) and verified them using SAS Version 9.3 (SAS Institute, Cary, NC) [10]. We used the Comprehensive Meta-Analysis software version 2 (Biostat, Englewood, NJ) [11] to verify results and assess comparable patterns in the data.

We multiplied our derived estimates of the population proportion of heterosexuals by the population estimate from the Census Bureau for persons aged 13 years or older for the 50 states and District of Columbia [12] to obtain an estimated number of heterosexuals. The population proportions of heterosexuals in the age group 15–24 years were applied to the population aged 13–24 years and the population proportions of heterosexuals in the age group 55–69 years were applied to the population aged 55 years or older. Because persons in the youngest age group (13–14 years) are less likely to have had sex than those aged 15–24 years [13], this may result in an over-estimate of the number of heterosexuals in the youngest age group and result in an under-estimate of the rates of HIV infection. It should be noted, however, that persons aged 13–14 years make up a relatively small proportion of the entire age group (13–24 years) and thus the overall effect of over-estimating the number of heterosexuals is likely to be small.

We calculated HIV rates by dividing the estimated number of HIV cases attributed to heterosexual contact (numerator) by the estimated number of heterosexuals (denominator). Corresponding to measures included in annual HIV surveillance reports [1], we calculated two types of HIV rates: 1) diagnosis rates and 2) the rates of heterosexuals living with diagnosed HIV infection. For the numerators, we used HIV case surveillance data from all 50 states and the District of Columbia reported to CDC as of June 2014 for adults and adolescents (age 13 years or older at diagnosis) diagnosed with HIV infection in 2013 and for those living with diagnosed HIV infection as of December 2012.

For the denominators, we used the estimated number of heterosexuals. Denominators were calculated by multiplying census data by the population proportion of heterosexuals derived from the meta-analysis. We used 2013 and 2012 census data, respectively, to determine the number of heterosexuals for the HIV diagnosis rates and rates of living with diagnosed HIV infection.

We calculated rate ratios to compare rates by sex, race/ethnicity, and age. Males, whites, and the youngest age group (13–24 years) served as the reference groups, respectively.

This study’s objective was to provide an estimate of the total number of heterosexuals in the U.S. and the HIV prevalence rate among heterosexuals, without accounting for the level of risk in their sexual behavior. Defining “high-risk heterosexuals” is complex [14]. Many people do not know the HIV status of their partners [14] and thus could not reliably report this risk. Other risk behaviors, such as high number of sex partners, are not sensitive enough to capture those who may have had only a single—albeit HIV-infected—partner. To calculate disease rates, the persons included in the numerator (HIV infection attributable to heterosexual contact) also must be included in the denominator (at risk for heterosexually acquired HIV infection). Definitions of “high-risk heterosexual” can exclude some HIV-infected persons in the denominator. Conversely, broader markers for HIV risk, such as any condomless sex, are not very specific and would not exclude those at low risk for infection in a population with comparatively lower rates of HIV prevalence.

We calculated the population proportion high-risk heterosexual and used that to calculate HIV prevalence. For the population proportion high-risk heterosexual, we used the data from NSFG [15] which reported approximately 5.6% of the general U.S. population age 15–44 years had a lifetime HIV-related sexual risk behavior, such as having 5 or more sex partners, exchanging sex for money or drugs, having a male sex partner who had sex with men or injected drugs, or had a sex partner who was HIV-positive. Subtracting the proportion of males that had male-male sex (2.1%) [15] gives a resulting estimate of 4.5% of the population being high-risk heterosexual.

Similar to our main analyses, we calculated HIV rates among high-risk heterosexuals by dividing the estimated number of HIV cases attributed to heterosexual contact (numerator) by the estimated number of high-risk heterosexuals (denominator). Denominators were calculated by multiplying census data by the 4.5% population proportion of high-risk heterosexuals derived from the NSFG data. The high-risk subset of the heterosexual population yields a smaller population denominator, thus the HIV prevalence of heterosexually acquired HIV infection calculated with the high-risk heterosexual estimate represents an upper bound estimate of HIV prevalence.

Table 2 shows the estimated population proportion of lifetime heterosexuals overall and for males and females for each population-based survey and the combined estimates from the meta-analysis. The overall combined estimate was 86.7% (95% confidence interval [CI]: 84.1%-89.3%). As noted, Q statistics and I² indicated heterogeneity of results across the surveys (I² = 91.6; Q = 23.7, p < 0.001). The combined estimate for males was 84.1% (CI: 81.2%-86.9%) and for females was 89.4% (95% CI: 86.9%-91.8%).

Applying these proportions to the U.S. population age 13 years or older for 2013, we estimate that approximately 228,402,110 adults and adolescents are heterosexuals, with an estimated range, based on the confidence intervals for the population proportion estimate, from 221,593,250 to 235,210,969 persons; using the sex-specific proportions represents an estimated 108,187,901 heterosexual males (range: 104,486,382–111,889,420) and 120,368,338 heterosexual females (range: 117,061,991–123,674,686). The proportion heterosexual among males is lower than among females in part because of exclusions for MSM and for PWID (the proportion PWID is higher among males than females) [3].

We calculated population proportion estimates for male and female heterosexuals by race/ethnicity and by age group (Table 3). The population proportion of heterosexuals did not differ significantly by race/ethnicity among males or females. The population proportion of heterosexuals was lowest among both males and females aged 15–24 years, and was not significantly different among those aged 25–69 years. Estimates for females for the age categories of 35–44 years, 45–54 years, and 55–69 years and the overall estimate for those aged 55–69 years had a relative standard error (RSE) of 30%-49%. In general, estimates with a RSE of 30% or greater are subject to high sampling error and are recommended to be used with caution. Thus, the specified estimates with RSE ≥30% and the resulting rates presented in Table 3 should be interpreted with caution.

Rates of diagnosis of HIV infection among heterosexuals and rates of heterosexuals living with diagnosed HIV infection are presented in Tables 4 and 5, respectively. The rate of diagnosis of HIV infection was 5.2 per 100,000 heterosexuals (CI: 5.1–5.4); the rate of heterosexuals living with diagnosed HIV infection was 104 per 100,000 heterosexuals (CI: 101–108), or 0.1%. The rates for females were higher than those for males for diagnosis rate (rate ratio: 1.9, CI: 1.7–2.0) and for rate of persons living with diagnosed HIV infection (rate ratio: 2.0, CI: 1.9–2.2).

The rate ratios revealed disparities by race/ethnicity and by age. Comparing black males to white males, the estimated rate of diagnoses of HIV infection was 24–31 times as high (Table 4) and the estimated rate of living with diagnosed HIV infection was 26–34 times as high (Table 5); comparing Hispanic/Latino to white males, these rates were approximately 5–6 times as high, for both rates (Tables 4 and 5). Comparing black females to white females, the estimated rate of diagnosis of HIV infection was 22–25 times as high (Table 4) and the estimated rate of living with diagnosed HIV infection was 20–23 times as high (Table 5); comparing Hispanic/Latino to white females, these rates were approximately 4–5 times as high for both measures (Tables 4 and 5). Among males, the rate of diagnoses and living with diagnosed HIV was higher among older age groups than those 13–24 years. The population proportion heterosexual among females stratified by age group had some RSE ≥30%, and thus rates and rate ratios should be interpreted with caution (Tables 4 and 5).

The HIV prevalence among high-risk heterosexuals was 2.0% (data not shown), or 20 times as high as our overall prevalence of 0.1%. Considering only this high-risk subset of the heterosexual population yields a smaller population denominator (4.5% vs. 87%) and thus a higher HIV prevalence than our result among all heterosexuals.