Previous Exposure to HCV Among Persons Born During 1945-1965
Previous Exposure to HCV Among Persons Born During 1945-1965
We analyzed NHANES data collected from 1999 to 2008. NHANES is an annual nationally representative multistage, stratified probability cluster survey of the US civilian, noninstitutionalized population. Information on the survey design and implementation, including institutional review board approval and consent, is detailed in the survey documentation.
Anti-HCV testing is administered to NHANES participants aged 6 years or older. We restricted our analysis to adult participants born from 1945 to 1965 who were interviewed and provided serum samples for anti-HCV testing. Birth year was estimated by subtracting participant age at time of survey from estimated year in which participant was surveyed. Because NHANES does not release data on participant birth year or the actual year in which a participant was interviewed or examined, we estimated the earliest survey year for each participant according to the variable, "six month time period when the examination was performed: November 1 through April 30, May 1 through October 31." As an example, for the 1999–2000 survey cycle, participants examined from November 1 to April 30 were assigned an earliest survey year of 1999 and those examined from May 1 through October 30 were assigned to the 2000 survey year. We excluded participants without specimens for testing and those with indeterminate anti-HCV results from the final analytic sample.
The outcome measure was anti-HCV prevalence as determined by serologic testing. We chose anti-HCV status as an endpoint because HCV RNA testing was not performed for the 2003–2004 NHANES cycle, and we determined that combining data from all 10 years (1999–2008) was necessary to achieve sufficient subdomain sample sizes for improved precision and reliability of point estimates.
Specimens were tested for antibodies to HCV by repeated enzyme-linked immunosorbent assay (ELISA version 3.0, Ortho Diagnostic Systems Inc, Raritan, NJ). Reactive specimens were confirmed by recombinant immunoblot assay (RIBA version 3.0, Chiron Corporation, Emeryville, CA). Participants who tested positive by both ELISA and RIBA were categorized as anti-HCV–positive.
We examined the following independent variables as potential predictors or confounders of anti-HCV prevalence within the birth cohort: race/ethnicity, gender, country of birth, veteran status, marital status, educational attainment, family income, health insurance status, daily alcohol consumption within the past 12 months, age at first sexual intercourse, number of lifetime sexual partners, lifetime IDU (cocaine, heroin, and methamphetamine), history of blood transfusion before 1992, and ALT level. We categorized race/ethnicity as non-Hispanic White, non-Hispanic Black, Mexican American, and other. We categorized educational attainment as completed less than high school and completed high school or more; marital status as married or living with partner, divorced or separated or widowed, and never married; and family income as greater than 2 times federal poverty threshold, 1 to 2 times federal poverty threshold, and less than the federal poverty threshold. We defined elevated ALT as 40 or more international units per liter. For independent variables with 10% or more of observations with missing values, we created an "unknown" category to include those missing values as valid for analysis. Accordingly, we categorized alcohol consumption as 0 or 1, 2 or more, and unknown number of drinks per day within the past year; age at first sexual intercourse as 17 years or younger, 18 years or older, and unknown; number of lifetime sexual partners as 0 to 9, 10 to 19, 20 or more, and unknown; lifetime drug use as never, non–injection drug use, IDU, and unknown.
NHANES questions related to sexual behavior and history of IDU are restricted to adult participants younger than 60 years. Thus, all analyses involving these variables in the current study were similarly restricted.
We generated proportions and 95% confidence intervals (CIs) to describe the characteristics of the 1945–1965 birth cohort. We also produced estimates of anti-HCV prevalence in the birth cohort and by subgroups. We specified linear contrasts of estimates to test for statistical differences in characteristics between anti-HCV–positive participants and all participants, and to test for differences in anti-HCV prevalence between subgroups. We assessed statistical reliability of estimated proportions by evaluating relative standard errors (< 30%) and by ensuring that subdomains met NHANES minimum sample size requirements. We generated unadjusted odds ratios from univariate logistic regression models. We defined statistical significance as P value less than .05.
We developed a multivariate logistic regression model to identify independent risk factors associated with anti-HCV positivity within the birth cohort after we controlled for covariates. We specified an estimated full model by including all independent variables with a P value of less than .1 from the univariate analyses. Using a backward elimination procedure, we removed variables with the lowest observed partial F-statistic at a predetermined P value of less than .1. We simultaneously tested for 2-way interaction effects between race and IDU or gender and IDU, by using multiple partial F-tests. We assessed multicollinearity among covariates by review of diagnostic statistics including variance inflation factors (> 2.5), condition indices (> 15), and variance proportions (> 0.5; SAS version 9.3, SAS Institute, Cary, NC). In deciding whether to exclude a covariate because of collinearity, we also considered the relative importance of the covariate, its relationship with key variables such as IDU, and its contribution to the overall model fit. We used the Hosmer–Lemeshow goodness-of-fit test to evaluate the overall fit of the final model. Except as otherwise specified, we analyzed all data with SAS-callable SUDAAN to account for the complex survey design (version 10.0.1, Research Triangle Institute, Research Triangle Park, NC). We rescaled sample weights after combining data across multiple survey years. We estimated variance and standard errors by using the Taylor series (linearization) method.
Methods
We analyzed NHANES data collected from 1999 to 2008. NHANES is an annual nationally representative multistage, stratified probability cluster survey of the US civilian, noninstitutionalized population. Information on the survey design and implementation, including institutional review board approval and consent, is detailed in the survey documentation.
Anti-HCV testing is administered to NHANES participants aged 6 years or older. We restricted our analysis to adult participants born from 1945 to 1965 who were interviewed and provided serum samples for anti-HCV testing. Birth year was estimated by subtracting participant age at time of survey from estimated year in which participant was surveyed. Because NHANES does not release data on participant birth year or the actual year in which a participant was interviewed or examined, we estimated the earliest survey year for each participant according to the variable, "six month time period when the examination was performed: November 1 through April 30, May 1 through October 31." As an example, for the 1999–2000 survey cycle, participants examined from November 1 to April 30 were assigned an earliest survey year of 1999 and those examined from May 1 through October 30 were assigned to the 2000 survey year. We excluded participants without specimens for testing and those with indeterminate anti-HCV results from the final analytic sample.
Outcome Variable
The outcome measure was anti-HCV prevalence as determined by serologic testing. We chose anti-HCV status as an endpoint because HCV RNA testing was not performed for the 2003–2004 NHANES cycle, and we determined that combining data from all 10 years (1999–2008) was necessary to achieve sufficient subdomain sample sizes for improved precision and reliability of point estimates.
Specimens were tested for antibodies to HCV by repeated enzyme-linked immunosorbent assay (ELISA version 3.0, Ortho Diagnostic Systems Inc, Raritan, NJ). Reactive specimens were confirmed by recombinant immunoblot assay (RIBA version 3.0, Chiron Corporation, Emeryville, CA). Participants who tested positive by both ELISA and RIBA were categorized as anti-HCV–positive.
Independent Variables
We examined the following independent variables as potential predictors or confounders of anti-HCV prevalence within the birth cohort: race/ethnicity, gender, country of birth, veteran status, marital status, educational attainment, family income, health insurance status, daily alcohol consumption within the past 12 months, age at first sexual intercourse, number of lifetime sexual partners, lifetime IDU (cocaine, heroin, and methamphetamine), history of blood transfusion before 1992, and ALT level. We categorized race/ethnicity as non-Hispanic White, non-Hispanic Black, Mexican American, and other. We categorized educational attainment as completed less than high school and completed high school or more; marital status as married or living with partner, divorced or separated or widowed, and never married; and family income as greater than 2 times federal poverty threshold, 1 to 2 times federal poverty threshold, and less than the federal poverty threshold. We defined elevated ALT as 40 or more international units per liter. For independent variables with 10% or more of observations with missing values, we created an "unknown" category to include those missing values as valid for analysis. Accordingly, we categorized alcohol consumption as 0 or 1, 2 or more, and unknown number of drinks per day within the past year; age at first sexual intercourse as 17 years or younger, 18 years or older, and unknown; number of lifetime sexual partners as 0 to 9, 10 to 19, 20 or more, and unknown; lifetime drug use as never, non–injection drug use, IDU, and unknown.
NHANES questions related to sexual behavior and history of IDU are restricted to adult participants younger than 60 years. Thus, all analyses involving these variables in the current study were similarly restricted.
Statistical Analysis
We generated proportions and 95% confidence intervals (CIs) to describe the characteristics of the 1945–1965 birth cohort. We also produced estimates of anti-HCV prevalence in the birth cohort and by subgroups. We specified linear contrasts of estimates to test for statistical differences in characteristics between anti-HCV–positive participants and all participants, and to test for differences in anti-HCV prevalence between subgroups. We assessed statistical reliability of estimated proportions by evaluating relative standard errors (< 30%) and by ensuring that subdomains met NHANES minimum sample size requirements. We generated unadjusted odds ratios from univariate logistic regression models. We defined statistical significance as P value less than .05.
We developed a multivariate logistic regression model to identify independent risk factors associated with anti-HCV positivity within the birth cohort after we controlled for covariates. We specified an estimated full model by including all independent variables with a P value of less than .1 from the univariate analyses. Using a backward elimination procedure, we removed variables with the lowest observed partial F-statistic at a predetermined P value of less than .1. We simultaneously tested for 2-way interaction effects between race and IDU or gender and IDU, by using multiple partial F-tests. We assessed multicollinearity among covariates by review of diagnostic statistics including variance inflation factors (> 2.5), condition indices (> 15), and variance proportions (> 0.5; SAS version 9.3, SAS Institute, Cary, NC). In deciding whether to exclude a covariate because of collinearity, we also considered the relative importance of the covariate, its relationship with key variables such as IDU, and its contribution to the overall model fit. We used the Hosmer–Lemeshow goodness-of-fit test to evaluate the overall fit of the final model. Except as otherwise specified, we analyzed all data with SAS-callable SUDAAN to account for the complex survey design (version 10.0.1, Research Triangle Institute, Research Triangle Park, NC). We rescaled sample weights after combining data across multiple survey years. We estimated variance and standard errors by using the Taylor series (linearization) method.
