Age and follicle-stimulating hormone as variables independently associated with serum anti-müllerian hormone in women attending a tertiary gynaecology clinic: A retrospective cross-sectional study using censoring-aware modelling

Objective: Anti-Müllerian hormone (AMH) is the most widely used biochemical marker of ovarian reserve, but its associations with other reproductive hormones are usually examined one hormone at a time and without accounting for values reported at the assay floor. This study examined the associations between serum AMH and age, follicle-stimulating hormone (FSH), luteinising hormone (LH), estradiol, progesterone and prolactin in women attending a tertiary gynaecology clinic. Materials and Methods: In this retrospective, cross-sectional, single-centre study, the records of women who underwent serum AMH testing together with a reproductive hormone panel between 1 January 2020 and 31 December 2025 were reviewed. Women with polycystic ovary syndrome, primary ovarian insufficiency, pregnancy or previous ovarian surgery were excluded. Bivariate associations were assessed with Pearson and Spearman correlation coefficients on raw and Box–Cox-transformed variables. Because 12.82% (n = 15) of AMH results were left-censored at the analytical reporting floor of 0.02 ng/mL, a multivariable Tobit model with the censoring limit specified on the Box–Cox-transformed scale (λ = 0.309, threshold = −2.270) was fitted, with predictors selected a priori on clinical grounds. Sensitivity analyses examined the influence of high AMH values, an alternative transformation of AMH, and flexible modelling of age. Results: The analysis included 117 women aged 18–45 years. Mean AMH was 2.22 ± 2.24 ng/mL [median 1.61 ng/mL (Q1–Q3, 0.36–3.20)]. In bivariate correlation analysis, AMH showed statistically significant inverse associations with age (r = −0.376, p < 0.001) and FSH (r = −0.445, p < 0.001), whereas LH, estradiol, progesterone, and prolactin showed no statistically significant correlations with AMH (all p > 0.10). In the multivariable Tobit model, only age (β = −0.068; 95% CI [−0.117, −0.018]; p = 0.007) and transformed FSH (β = −3.582; 95% CI [−4.846, −2.317]; p < 0.001) remained independently associated with transformed AMH. LH, estradiol, progesterone, and prolactin were not independently associated with AMH in the multivariable model (all p > 0.05). The standardised association was greater for transformed FSH (β* = −0.559) than for age (β* = −0.234). The associations of age and FSH were consistent in direction and magnitude across all sensitivity analyses. Conclusions: In this selected clinical sample, age and FSH were the only variables independently associated with serum AMH after mutual adjustment. LH, estradiol, progesterone, and prolactin showed no statistically significant associations with AMH in either the unadjusted bivariate analyses or the multivariable model. The absence of independent associations for these hormones should be interpreted cautiously because hormone sampling was not standardised to menstrual cycle day, which may have introduced measurement variability. These findings describe associations within the study population and do not establish causal relationships, reference intervals, or a basis for modifying testing strategies. The analysis also highlights the importance of accounting for left-censoring when modelling AMH data containing a substantial proportion of results at the assay reporting floor.