Higher-Normal Thyroid Levels Linked to Lower IVF Pregnancy Odds in Cluster Study

Women undergoing in vitro fertilization whose thyroid-stimulating hormone sits at the upper end of the conventional normal range had significantly lower odds of achieving a clinical pregnancy than women in the middle of that range, according to a retrospective cohort study of 878 euthyroid patients treated at a single fertility center in Iraq. The finding, published in Frontiers in Endocrinology, challenges the widespread practice of treating the entire normal TSH reference interval as uniformly favorable and suggests that a single binary cutoff may obscure clinically meaningful gradations within it.

The research team, led by Mohamed M. Amin of the German University in Cairo together with SI. Hassanein, analyzed records from women who underwent fresh IVF or ICSI cycles at Al-Hayat Hospital’s infertility unit between January 2022 and December 2024. All participants were euthyroid, with TSH values between 0.25 and 4.0 micro-international units per milliliter, and all received transfers of Grade A embryos. Women with overt thyroid dysfunction, thyroid hormone treatment, autoimmune disease, or other major endocrine conditions were excluded, leaving a relatively homogeneous population in which thyroid status varied only within accepted normal limits.

Instead of dividing patients at an arbitrary threshold such as 2.5 micro-international units per milliliter, the investigators applied k-means clustering, an unsupervised machine-learning method, directly to the pretreatment TSH values. The algorithm partitioned the cohort into four data-driven groups: Cluster 1 with a mean TSH of about 3.47 and a range of 2.9 to 4.0, Cluster 2 with a mean of 2.50, Cluster 3 with a mean of 1.73 and a range of 1.4 to 2.0, and Cluster 4 with a mean of 0.92. Internal validation supported the stability of the four-cluster solution, with a silhouette coefficient of 0.549, a Calinski-Harabasz index of 3086.67, and complete concordance between the k-means partition and a hierarchical Ward-linkage clustering of the same data.

Clinical pregnancy, defined as the visualization of a gestational sac with a confirmed fetal heartbeat on ultrasound at six to seven weeks of gestation, was achieved by 474 of the 878 women, an overall rate of 54 percent. Pregnancy rates, however, were not evenly distributed across the TSH clusters. Cluster 3, the mid-normal group, had the highest rate at 58.2 percent, while Cluster 1, the highest-normal group, had the lowest at 42.7 percent. Cluster 4 reached 56.0 percent and Cluster 2 stood at 54.1 percent. A global chi-square test across clusters yielded a p-value of 0.0268, with the pairwise difference between Cluster 1 and Cluster 3 reaching p = 0.024.

The association persisted after statistical adjustment. In a stepwise logistic regression controlling for age, antral follicle count, number of stimulation ampoules, oocyte yield, oocyte maturation rate, and fertilization rate, women in Cluster 1 had 47 percent lower odds of clinical pregnancy than those in the reference Cluster 3, with an odds ratio of 0.53 and a 95 percent confidence interval of 0.35 to 0.80. Clusters 2 and 4 did not differ significantly from the reference group. When TSH was modeled as a continuous variable in a separate multivariate model, each unit increase was likewise associated with reduced odds, with an odds ratio of 0.84 and a 95 percent confidence interval of 0.72 to 0.98.

The regression analysis also confirmed the expected contributions of established reproductive predictors. Younger age, higher oocyte yield, greater oocyte maturation rate, and higher fertilization rate each favored pregnancy, while greater gonadotropin consumption, measured in ampoules of 75 international units, was inversely associated, reflecting the poorer ovarian responsiveness of patients who require more stimulation. Notably, baseline characteristics such as age, body mass index, FSH, LH, day-2 estradiol, endometrial thickness, antral follicle count, and stimulation duration did not differ significantly across the four TSH clusters, and fertilization rates were comparable across groups, indicating that thyroid status was the principal determinant of cluster separation rather than a proxy for ovarian reserve.

Subgroup analyses by infertility diagnosis added nuance. Unexplained infertility showed the steepest gradient, with pregnancy rates rising from 34.3 percent in Cluster 1 to 65.6 percent in Cluster 3, and ovarian factor infertility followed a similar pattern, climbing from 38.9 percent to 63.0 percent. Male factor infertility remained comparatively stable across clusters, at roughly 52 to 63 percent, while cases of polycystic ovary diverged from the overall trend, with the highest pregnancy rate, 66.6 percent, observed in Cluster 1, the highest-normal TSH group. The authors suggest that the complex endocrine environment of polycystic ovary syndrome, or differences in stimulation protocols for these patients, may modify any thyroid-related association, though the retrospective design cannot establish mechanisms.

The study’s authors and the broader literature both urge caution in interpretation. Prior research on TSH within the normal range has been conflicting: some studies reported lower clinical pregnancy rates above 2.5 micro-international units per milliliter, while others, including a large retrospective cohort and meta-analysis of more than 13,000 women and a random-forest analysis of 1,382 IVF cycles, found no meaningful effect of TSH on live birth or pregnancy prediction. The largest interventional evidence, the TABLET trial of 952 euthyroid women with thyroid autoimmunity, showed that levothyroxine supplementation did not improve live birth rates, underscoring that an observational association cannot be assumed to be treatable.

Several limitations constrain the conclusions. The retrospective, single-center design precludes causal inference and leaves open the possibility of unmeasured confounding; factors such as iodine status and thyroid peroxidase antibodies, which were not routinely tested at the center, could not be accounted for. The endpoint was clinical pregnancy rather than live birth, which remains the definitive measure of IVF success, and the population was restricted to women with TSH already within the normal range, so the results cannot be extended to overt thyroid dysfunction. The authors explicitly frame their findings as an association rather than evidence of causation.

Within those limits, the work makes a methodological point with potential clinical relevance: data-driven stratification of a continuous hormone variable can reveal patterns that binary cutoffs conceal. The authors argue that individualized thyroid assessment, informed by infertility context and ovarian response parameters, may be more appropriate than a universal threshold, and they call for prospective studies incorporating comprehensive thyroid panels, iodine measurement, and live birth outcomes to test whether the gradient they observed holds up under stronger designs. Until such evidence accumulates, the study stands as a provocative signal that the upper-normal TSH range, long treated as uniformly safe in fertility practice, may not be uniformly favorable.

Subject of Research: The association of TSH variation within the normal reference range with clinical pregnancy outcomes in IVF, analyzed using k-means clustering.

Article Title: Higher TSH within the normal range is independently associated with reduced IVF clinical pregnancy odds: a K-means cluster analysis challenging the binary cutoff paradigm

Article References: Amin, M. M., & Hassanein, S. (2026). Higher TSH within the normal range is independently associated with reduced IVF clinical pregnancy odds: a K-means cluster analysis challenging the binary cutoff paradigm. Frontiers in Endocrinology, 17, Article 1856685. https://doi.org/10.3389/fendo.2026.1856685

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1856685

Keywords: TSH, IVF, clinical pregnancy, k-means clustering, thyroid, infertility, retrospective cohort, euthyroid, assisted reproduction, odds ratio, Frontiers in Endocrinology, subfertility