The RYMS study population was quite homogeneous. Participants were 18–22 years of age (median age, 19.4 years), predominantly Caucasian (81%), nonsmokers (73%), with a median BMI of 24.2. Median sperm concentration was 53.5 × 106/mL, and median total sperm count was 157 × 106. Demographic and reproductive parameters are summarized in . Although this was a population of apparently healthy young men, 24.6% of them had a sperm count < 20 × 106/mL, a commonly used cutoff for subfertility (WHO 1999).
Characteristics of RYMS participants (n =
The distributions of both AGDAS and AGDAP were approximately normal (). AGDAS (mean, 51.3 mm; median, 51.7 mm) was, on average, 40% as long as AGDAP (mean, 128 mm; median, 126 mm), and the SDs of these two measures were similar (14.5 mm and 13.0 mm for AGDAS and AGDAP, respectively) (). As expected, AGDAS and AGDAP were highly correlated [Pearson correlation (R) = 0.60, p < 0.0001].
Frequency distributions of AGDAS (A) and
AGDAP (B) in our study population.
Variability of AGD measurements. Variation with period of examination. Here we refer to fall 2009 as the first recruitment period and spring 2010 as the second recruitment period. We observed a small but significant decrease in both AGDAS and AGDAP between the first recruitment period (n = 44) and the second (n = 82). Mean AGDAS was 56.6 mm and 48.5 mm, and mean AGDAP was 132 mm and 126 mm, for men recruited during period 1 and period 2, respectively. Mean age, BMI, abstinence time, and all semen parameters were similar in the two recruitment periods. Despite these differences between time periods, we saw no significant time trend within periods. Although neither exam date nor recruitment period was associated with any of the semen parameters (all p-values > 0.24), we retained recruitment period in all final models.
Within- and between-examiner variability. The mean (absolute) difference within examiners was 1.39 mm for AGDAS (2.7% of mean AGDAS) and 2.62 mm for AGDAP (2.1% of mean AGDAP). We used a mixed model to estimate the interclass correlations, which were 0.91 [95% confidence interval (CI), 0.79–0.97] and 0.95 (95% CI, 0.89–0.98) for AGDAS and AGDAP, respectively. We repeated all regression analyses omitting the seven subjects measured by J. Stevens, and results were unchanged.
Predictors of AGDAS and AGDAP. We examined variables that predicted AGDAS and AGDAP (). BMI, height, and recruitment period (fall 2009 vs. spring 2010) were significant predictors of both measures. Testicular volume, testicular abnormalities, stress, and ethnicity were not significantly related to AGD, so we did not retain them in the final models.
Predictors of AGDAS and
AGDAP in multivariate models.
The model fit was better for predicting AGDAP than AGDAS (adjusted R2 = 0.45 and 0.23, respectively). BMI accounted for most of the variability in AGDAP but little of the variability for AGDAS, whereas height was more influential for AGDAS.
AGD and other covariates in relation to semen parameters. Covariates retained in final models predicting semen parameters were AGD measures, height, recruitment period, ethnicity (African American or not), abstinence time, and time to sample analysis as described in “Materials and Methods.”
AGDAS was positively related to sperm concentration, motility, morphology, total sperm count, and total motile count (p-values 0.002, 0.028, 0.048, 0.006, and 0.009, respectively; ). The associations between AGDAP and sperm count and concentration were negligible, although in a similar direction as those for AGDAS. Regression coefficients for the two AGD measures as predictors of sperm motility and morphology were not inconsistent, although CIs for AGDAP were wider and consistent with no association. The residual plots for sperm concentration in relation to AGDAS and AGDAP from our multivariate models, are shown in .
Multivariate analysis for men’s semen parameters
and AGDAS and AGDAP.a
Partial regression plot (mean ± SE) of sperm concentration
modeled as a function of (A) AGDAS and (B)
We also examined sperm concentration dichotomized at 20 × 106/mL (subfertile vs. normal) in relation to AGD, controlling for the same covariates used in the linear regression models. AGDAS was significantly related to this outcome. The risk of subfertility was increased 7.3 times (95% CI, 2.5–21.6) for an (adjusted) AGDAS below the median, compared with AGDAS above the median. Having a low sperm concentration (< 20 × 106/mL) was inversely related to AGDAS (p < 0.0019). AGDAP was not related to this outcome.
We calculated the expected change in semen parameters associated with an interquartile increase in AGDAS for a typical study participant. When AGDAS is 43.1 mm, the 25th percentile of the AGD distribution, the expected sperm concentration, using our final regression model, is 34.7 × 106/mL. When AGDAS is 61.1 mm (the 75th percentile), the expected sperm concentration is 51.6 × 106/mL, whereas the predicted value for the 50th percentile of AGDAS is 42.0 × 106/mL. Thus, an interquartile increase in AGDAS is associated with an increase in sperm concentration that is 40.2% of the median, based on the best-fitting model. Similar increases are seen for other sperm parameters, although a smaller increase is seen for percent morphologically normal sperm.
Height, BMI, and time period were all associated with AGD and included in the final models, although none of these variables was associated with any semen parameter in this population. All sperm parameters were significantly lower in the small subgroup (n = 7) of African-American men compared with other men in this population (p-values for sperm parameters, < 0.001 to 0.016).