Data on ANA were available for 13,519 participants from five NHANES cycles: 1988-1991, 1999-2000, 2001-2002, 2003-2004, and 2011-2012. The NHANES sampled representative members of the noninstitutionalized civilian US population and provided sampling weights to adjust for selection probabilities and nonresponse (41), which enables inference that generalizes to most of the US population. All participants signed informed consent documents and completed questionnaires, and most were physically examined and provided blood and urine specimens. Available data included demographic characteristics, health covariates, measured factors, and constructed variables. The NHANES protocol was approved by the Human Subjects Institutional Review Board of the US Centers for Disease Control and Prevention (CDC).

All serum samples were evaluated for ANA in the laboratory of Dr. Edward K.L. Chan between 2016 and 2017 by indirect immunofluorescence at a 1:80 dilution using the NOVA Lite HEp-2 ANA slide with DAPI kit (Inova Diagnostics, San Diego, California, USA), with a highly specific fluorescein isothiocyanate-conjugated secondary antibody (goat anti-human IgG). Immunofluorescence staining intensities were graded 0–4 compared to standard references (42). Grades 1–4 were considered positive for ANA and grade 0 was considered negative. For more assay details see Dinse et al. (7).

For cost and other practical reasons, ANA were only assayed in a subset of participants ≥12 years old in each of the five cycles. The ANA subsamples from 1999-2000, 2001-2002, and 2003–2004 were each roughly one-third the size of those from 1988–1991 and 2011-2012. Thus, as in our earlier studies (7, 8), we combined the three middle cycles to create three time periods with similar sample sizes: 1988-1991 (N=4,727), 1999-2004 (N=4,527), and 2011-2012 (N=4,265). As before, we focused on these three periods rather than the five cycles.

All analyses were restricted to the 13,519 participants with ANA data. The CDC adjusted the sampling weights to account for analyzing this ANA subsample. Data were available on cotinine, and thus smoking exposure as defined by cotinine concentration, for 13,288 participants; on self-reported smoking history for 12,278 participants; and on both smoking exposure and smoking history for 12,063 participants. Supplementary Table S1 shows the numbers of participants in each time period (and overall) with data on ANA, smoking exposure, and smoking history. Throughout this article, “cotinine” refers to serum cotinine and not urinary cotinine.

The ANA outcome variable was a binary indicator of ANA positivity/negativity. Cotinine concentration (ng/mL) was a quantitative variable and was used to classify smoking exposure as none (≤0.05), passive (>0.05 to 10), or active (>10), as recommended by the CDC and the US Environmental Protection Agency (EPA) (13), though a sensitivity analysis applied a more recent recommendation of >3 ng/mL for defining active smoking exposure. Combining the first two exposure categories produced an indicator of smoking status: nonsmoker (none or passive exposure) versus smoker (active exposure). Smoking history was based on questionnaire data, with individuals self-reporting as never, former, or current smokers.

Except where otherwise noted, our primary analyses adjusted for sex, age, race/ethnicity, and the survey design variables (i.e., strata, clusters, and weights proportional to the inverse probability of sampling), each of which was available for all participants. Age was measured in years and categorized by decade (12-19, 20-29,…, 70-79, ≥80), though sensitivity analyses explored the use of fewer age categories, a quantitative age variable, or a restricted cubic spline in age. Self-reported race/ethnicity was categorized as non-Hispanic White, non-Hispanic Black, Mexican American, or Other. Secondary analyses adjusted for body mass index (BMI), alcohol intake, poverty income ratio (PIR), and education, as defined previously (42). Secondary analyses also investigated CO content in cigarettes, pack-years of smoking, lifetime years of smoking, and years since former smokers quit smoking, though these data were very limited.

When analyzing ANA prevalence, we used logistic regression models to allow the probability of ANA positivity to depend on explanatory variables. All models adjusted for the survey design variables. The basic model for estimating overall ANA prevalence and its 95% confidence interval (CI) did not include adjustment covariates, but we did include a categorical covariate for period when estimating ANA prevalence in each of the three time periods. When assessing ANA time trends, we adjusted for sex, age, and race/ethnicity and calculated an ANA prevalence odds ratio (OR) and its 95% CI for each period relative to the first period. The statistical significance of an ANA time trend was evaluated by replacing the categorical period covariate with a quantitative time covariate and then inspecting its p-value, where time was defined as the number of years between the midpoints of the participant’s period and the first period.

When analyzing the cotinine data, we calculated the geometric mean cotinine concentration for each time period. We also derived a trend line by using linear regression to model individual log-transformed cotinine concentration as a function of the number of years between the midpoints of the participant’s period and the first period. Any concentration below the limit of detection (LOD) was replaced by an imputed value of LOD/ 2 (43, 44). The cotinine LOD was initially 0.05 ng/mL but was lowered to 0.015 ng/mL during the second period due to an improvement in the assay; the corresponding imputed values were 0.035 and 0.011 ng/mL. We also evaluated mean cotinine concentrations over time (and estimated trend lines) within subgroups of self-reported never, former, and current smokers, and we used kernel density plots to assess the full cotinine concentration distribution for each smoking-history subgroup and time period.

When analyzing smoking time trends, we used logistic regression to estimate the prevalence of smokers in each time period. Overall prevalence estimates were adjusted for the survey design variables but not for any covariates. Also, after further adjusting for sex, age, and race/ethnicity, we estimated a prevalence OR (and a 95% CI) for each period relative to the first period.

When investigating the relationship between ANA and smoking, we performed logistic regression analyses similar to those described above for ANA prevalence. First, we stratified by smoking and analyzed ANA prevalence and time trends separately in each stratum. Second, we stratified by both age and smoking to see whether the ANA association with smoking depended on age. Third, we added a smoking covariate (instead of stratifying) and assessed whether that smoking covariate affected the ANA association with time or whether removing the period covariate altered the ANA association with smoking. Fourth, we also added a smoking-by-period interaction to evaluate whether smoking modified the ANA time trend. Fifth, we stratified by period and compared ANA prevalence for smokers versus nonsmokers to gauge how the ANA association with smoking changed over time.

Finally, we conducted sensitivity analyses to assess whether our results changed when using an alternative age covariate (fewer categories, quantitative, or restricted cubic spline) or when only considering adults (ages ≥20 years). We also explored the use of a more recent recommendation of >3 ng/mL for the cotinine cutpoint when defining active smoking exposure. In addition, we investigated several other covariates (BMI, alcohol intake, PIR, and education) and the limited data on cigarette CO content, pack-years, years of smoking, and years since quitting.

All analyses were performed with SAS software (version 9.4, SAS Institute, Cary, NC) and accounted for the survey design variables by using special survey procedures. Domain statements were used to properly handle the sampling weights in subgroup analyses. Variance estimates for the 95% CIs were obtained using the Taylor series method. Reported p-values were 2-sided. All plots were constructed in SAS except the kernel density plot, which was created in R (version 4.4.0, R Foundation, Vienna, Austria).

The prevalence of ANA rose over the 25-year span for which NHANES data on ANA were available, with most of the increase occurring between the second and third time periods. Accounting only for time period and the survey design variables, the weighted estimates of ANA prevalence were 11.0% (95% CI: 9.7-12.5%) in Period 1 (1988–1991), 11.4% (95% CI: 10.2-12.8%) in Period 2 (1999-2004), and 16.1% (95% CI: 14.5-17.9%) in Period 3 (2011-2012). These overall estimates, along with sample sizes and numbers of ANA-positive participants, are shown in the last row of Table 1 . Relative to Period 1 and after further adjustment for sex, age, and race/ethnicity, the ANA prevalence OR was 1.02 (95% CI: 0.84-1.24) for Period 2 and 1.49 (95% CI: 1.23-1.82) for Period 3 ( Table 2 ), and there was strong statistical evidence of a positive trend in ANA prevalence over time (p=0.0001). We reported these results earlier (7), with slight discrepancies due to minor differences in analysis, but repeat them here for context.

Supplementary Figure S1 shows the geometric mean cotinine concentration and its 95% CI for each period, along with the best-fitting trend line. There was strong statistical evidence (p<0.0001) of a steady decrease over time. When stratified by self-reported smoking history, the mean cotinine levels ranged from 0.04 to 0.27 ng/mL for never smokers, 0.08 to 0.59 ng/mL for former smokers, and 104.2 to 158.5 ng/mL for current smokers (top half of Supplementary Table S2 ). The best-fitting trend line had a negative slope in all three subgroups but was steeper for never and former smokers than for current smokers ( Figure 1 ). Also, the decrease over time was statistically significant for both never and former smokers (p<0.0001), but not for current smokers (p=0.08). Thus, on average, current smokers had cotinine levels that were high and fairly constant over time, while former and never smokers had levels that were low and decreasing, likely due to steady reductions in secondhand smoke exposure. Similar results were obtained when excluding participants under age 20 years (bottom half of Supplementary Table S2 ) to account for smoking history data being available for different age ranges across time periods (≥17 years in Period 1, ≥12 years in Period 2, and ≥20 years in Period 3).

Rather than focusing on means, Figure 2 displays kernel density estimates of the entire distribution of cotinine concentrations by time period and smoking history. These plots clearly show the differences in cotinine levels for never and former smokers (low) versus current smokers (high), as well as the consistency over time for current smokers. The cotinine distributions for never and former smokers were less consistent, with a notable shift toward lower values as time progressed. Much of this shift was likely due to many never and former smokers having cotinine levels below the LOD, which decreased from 0.05 to 0.015 ng/mL in the second time period. Nondetectable levels were replaced by imputed values of 0.035 and 0.011 ng/mL, respectively, which match well with the peaks of the period-specific cotinine distributions for never smokers. The cotinine distributions were more spread out for former smokers than for never smokers, perhaps due to a larger proportion of former smokers interacting with a current smoker.

Cigarette smoking in the US has decreased for a half-century (16, 17). We confirmed this downward trend in the NHANES data by examining the proportions of active smokers (defined by cotinine levels) and current smokers (based on self-reports), both of which clearly decreased over time. Unadjusted period-specific estimates of smoking prevalence for both classifications demonstrated similar decreases among all participants and among adults only ( Supplementary Table S3 ), as did covariate-adjusted estimates of the smoking prevalence ORs for time period ( Supplementary Table S4 ).

Estimates of ANA prevalence exhibited different temporal patterns in the three smoking exposure subgroups. For individuals with no exposure, these estimates were highest but did not show a clear trend; for passive exposure, they were intermediate and increased steadily across all periods from 11.1% (95% CI: 9.6-12.8%) to 12.7% (95% CI: 10.5-15.4%) to 15.5% (95% CI: 13.2-18.1%); and for active exposure, they were lowest and initially flat but then rose markedly from 7.4% (95% CI: 5.6-9.7%) in Period 2 to 13.3% (95% CI: 11.0-15.9%) in Period 3 ( Table 1 ). Covariate-adjusted estimates of the ANA prevalence OR for Period 3 relative to Period 1 were 0.99 (95% CI: 0.62-1.57) for no exposure, 1.72 (95% CI: 1.33-2.22) for passive exposure, and 1.45 (95% CI: 1.01-2.08) for active exposure ( Table 2 ). When assessing a linear trend in ANA prevalence across all three periods, the p-values for the three exposure subgroups were 0.2139, 0.0001, and 0.0661, respectively ( Table 2 ). We reported similar estimates previously (7), but with smoking exposure categories defined by slightly different cutpoints for cotinine concentration.

To investigate whether age modified the association between smoking and temporal patterns of ANA, in addition to stratifying by smoking exposure, we further stratified by three age groups (12-19, 20-49, and ≥50 years) instead of including a categorical covariate for age. This approach essentially allowed for interactions between age and the covariates (sex, race/ethnicity, and time period). Despite the larger number of subgroups leading to smaller counts within each, there was statistical evidence that the observed increase in ANA prevalence over time was associated mainly with 12–19 year-olds who were passive (p=0.005) or active (p=0.003) smokers ( Table 3 ). Among adolescents and relative to Period 1, the ORs and 95% CIs for passive smokers were 1.63 (0.83-3.23) in Period 2 and 2.64 (1.37-5.08) in Period 3, and for active smokers they were 3.01 (0.53-17.3) in Period 2 and 9.92 (2.20-44.7) in Period 3. The wide CIs are indicative of the small counts, but the ORs are large, especially for adolescents who were active smokers (which would have included vapers), for whom the odds of being ANA positive were roughly 10 times greater in Period 3 compared with Period 1. The differences across age categories, based on assessing an interaction between age group and time period, were statistically significant (p=0.009).

In an alternative covariate-adjusted analysis, we focused on smoking status and assessed the odds of ANA positivity for smokers relative to nonsmokers ( Table 4 ). Overall, smokers were less likely to have ANA than nonsmokers (OR=0.73; 95% CI: 0.58-0.92; p=0.007). When stratified by time period, the ANA prevalence ORs for smoking status varied in magnitude and statistical significance but not in direction. The odds of having ANA were significantly lower for smokers than nonsmokers in Period 2 (OR=0.65; 95% CI: 0.45-0.93; p=0.020), but that inverse association was weaker and not statistically significant in Period 1 (OR=0.80; 95% CI: 0.52-1.22; p=0.297) and Period 3 (OR=0.82; 95% CI: 0.56-1.21; p=0.310). This nonmonotonic temporal pattern is illustrated in Figure 3 , where ANA prevalence estimates are smaller for smokers than nonsmokers in all three time periods, but the difference is much greater in Period 2 than in Periods 1 and 3.

We performed several sensitivity analyses by adding covariates to a base model that was adjusted for sex, age, race/ethnicity, and time period ( Supplementary Table S5 ). Rather than stratifying by smoking exposure, including it as a categorical covariate led to the same basic pattern of ANA prevalence not changing much between Periods 1 and 2, followed by a marked increase in Period 3. When we also added a smoking-by-period interaction, the main effects of both smoking and period were statistically significant, but the interaction was not. On the other hand, excluding time period significantly worsened the model fit (p<0.0001), suggesting that calendar time was important and that smoking on its own could not fully explain the observed ANA differences.

We also performed secondary analyses that accounted for BMI, which had been shown previously to modify ANA time trends (7). Adding a 3-level categorical covariate for BMI (underweight/healthy, <25; overweight, 25 to <30; or obese, ≥30) to the base model did not change the ANA time trends, nor did also adding a BMI interaction with any factor in the base model or in an expanded model that also included a covariate for smoking exposure ( Supplementary Table S5 ). Similarly, the original ANA time trends within smoking exposure subgroups (as shown in Table 2 ) did not change much when augmenting the base model with covariates for a BMI main effect and a BMI-by-age interaction ( Supplementary Table S6 ).

Replacing the 8-category age covariate with a 3-category age covariate, a quantitative age covariate, or a restricted cubic spline in age did not alter the basic pattern of ANA prevalence being relatively flat between Periods 1 and 2, and then increasing substantially in Period 3 ( Supplementary Table S5 ). Similarly, that basic ANA pattern also remained consistent when other covariates were added to the base model, such as an individual main effect for alcohol intake, PIR, or education; those same main effects plus a main effect for BMI; and those same main effects plus both a main effect for BMI and an interaction between BMI and each of those other covariates.

Data were available on the CO content in the brand of cigarettes used by each of 1,157 current smokers from Periods 2 and 3 aged ≥20 years. After adjusting for sex, age, race/ethnicity, and time period, there was mild evidence that ANA prevalence decreased as CO content per cigarette increased (OR=0.92; 95% CI: 0.85-1.00; p=0.042). We also multiplied CO content by average number of cigarettes smoked per day to estimate total CO, but the covariate-adjusted analysis showed no evidence of an association with ANA (p=0.65). In additional covariate-adjusted analyses of participants of all ages from all time periods, we found no evidence of an ANA association with pack-years among 4,795 ever (former or current) smokers (p=0.91), years of smoking among 5,047 ever smokers (p=0.18), or years since quitting among 2,738 former smokers (p=0.56). We also allowed for various pack-year threshold values, but no ANA associations with pack-years were significant.