Optimal androgen concentration cut-off values for identifying abnormal insulin sensitivity in Korean women with polycystic ovary syndrome
Article information
Abstract
Objective
This study evaluated the relationship between serum androgen concentrations and validated insulin sensitivity indices in Korean women with polycystic ovary syndrome (PCOS) and established optimal androgen thresholds for identifying abnormal insulin sensitivity.
Methods
A total of 180 Korean females aged 18–35 years with newly diagnosed PCOS were enrolled. Fasting-state insulin sensitivity assessment indices (ISAIs) were calculated and participants with abnormal levels according to at least one established ISAI criterion were classified as having abnormal insulin sensitivity. Spearman’s correlation analysis assessed relationships between insulin sensitivity parameters and four androgens: total testosterone (TT), free testosterone (FT), dehydroepiandrosterone sulfate (DHEA-S), and 17α-hydroxyprogesterone (17-OHP). Receiver operating characteristic (ROC) curve analysis was used to define the most accurate androgen cutoff values for predicting abnormal insulin sensitivity.
Results
Among the four androgens tested, TT, FT, and DHEA-S showed significant associations with most insulin resistance (IR) parameters, whereas 17-OHP showed no significant correlation. In the ROC curve analysis, only FT achieved moderately accurate predictive performance (area under the curve [AUC], 0.759), while TT, DHEA-S, and 17-OHP showed AUC values indicating less accurate predictions. The optimal FT cutoff to detect abnormal insulin sensitivity was 1.325 pg/mL (sensitivity, 77.8%; specificity, 64.0%).
Conclusion
FT showed the strongest association with parameters related to insulin sensitivity and could serve as a complementary opportunistic indicator of IR in PCOS. Women with PCOS who had serum FT levels ≥1.325 pg/mL were identified as having abnormal insulin sensitivity.
Introduction
Approximately 6–13% of reproductive-aged women have polycystic ovary syndrome (PCOS), a major endocrine disorder in this population [1,2]. The prevalence varies according to the diagnostic criteria applied [3]. This syndrome encompasses reproductive, metabolic, and psychological manifestations and is characterized by oligo/anovulation, clinical and/or biochemical hyperandrogenism (HA), and polycystic ovarian morphology (PCOM) on ultrasound [4]. Beyond its reproductive implications, PCOS has been increasingly recognized as a complex systemic metabolic disorder in which insulin resistance (IR) and compensatory hyperinsulinemia play central pathophysiologic roles [1].
Traditionally, most studies have focused on the mechanisms by which IR and hyperinsulinemia contribute to HA [5]. Insulin acts synergistically with luteinizing hormone to enhance cytochrome P450c17α activity in ovarian theca cells, thereby stimulating androgen biosynthesis [6], and concurrently suppresses hepatic synthesis of sex hormone-binding globulin (SHBG), which increases circulating free testosterone (FT) concentrations [7]. These complementary mechanisms establish a well-recognized pathophysiological link between IR and HA, suggesting that hyperinsulinemia plays a central role in excess ovarian androgens in women with PCOS [5–8].
Recent studies have suggested that this relationship may be bidirectional [8]. Animal models and clinical studies suggest that excess androgens can induce or aggravate IR [8,9]. Androgens impair insulin signaling by limiting glucose uptake in muscle and fat and by suppressing glucose transporter (GLUT-4) expression [10]. Moreover, excess androgen promotes visceral fat accumulation and impairs hepatic insulin clearance, further deteriorating insulin sensitivity [11]. This bidirectional interplay between HA and IR forms a potentially self-reinforcing pathophysiological loop in the pathogenesis of PCOS; however, longitudinal studies confirming temporal causality remain limited.
Despite this understanding, clinical studies evaluating the association between androgen concentrations and IR biomarkers remain scarce [12,13]. Most studies have focused on the correlation between individual androgens and IR indices; however, standardized cutoff values for clinical use are lacking. Identifying the optimal androgen thresholds associated with an increased metabolic risk could have meaningful clinical implications. Since serum androgens are routinely measured during standard PCOS evaluations, an FT-based threshold could serve as an opportunistic metabolic indicator, enabling the simultaneous assessment of androgen excess and early insulin dysregulation without requiring additional laboratory tests. This is particularly relevant for identifying women with a subclinical metabolic risk before overt glucose dysregulation manifests.
In 2023, we reported that fasting insulin levels could effectively predict HA in women with PCOS, with an optimal cutoff of 9.85 μU/mL (area under the curve [AUC], 0.817) [14]. This suggests that insulin dysregulation drives excess androgen production. Building on this foundation, the present study aimed to explore the reverse pathway of this relationship, that is, whether elevated androgen concentrations can identify impaired insulin sensitivity in the same clinical population. Specifically, we sought to 1) comprehensively evaluate the associations between serum concentrations of four androgens (total testosterone [TT], FT, dehydroepiandrosterone sulfate [DHEA-S], and 17α-hydroxyprogesterone [17-OHP]) and multiple validated indices of insulin sensitivity/resistance; 2) determine the optimal androgen cutoff values to detect impaired insulin sensitivity in women with PCOS through receiver operating characteristic (ROC) curve analysis; and 3) compare the predictive performance of different androgens to identify the most informative biomarker for metabolic risk assessment.
Materials and methods
1. Subjects
This retrospective cross-sectional study was conducted at a Inje university Haeundae Paik Hospital. The study period was from January 2010 to December 2013. We included Korean women aged 18–35 years who were newly diagnosed with PCOS during this period. All study participants were >8 years post-menarche at the time of enrollment. PCOS was diagnosed using the Rotterdam criteria [4], excluding other etiologies (congenital adrenal hyperplasia, androgen-secreting tumors, and Cushing’s syndrome). Clinical HA was defined by the presence of hirsutism (modified Ferriman-Gallwey score ≥6) [4]. Biochemical HA was defined as an elevated serum androgen level beyond the 95% confidence limits defined in controls by Chae et al. [15] (FT >1.72 pg/mL and/or TT >0.68 ng/mL). Women were also excluded if they had diabetes, thyroid disorders, or hyperprolactinemia diagnosed before enrollment; had undergone ovarian surgery; or had recent exposure (within 6 months) to medications altering sex hormone or gonadotropin levels, including oral contraceptives, ovulation induction drugs, glucocorticoids, or antiandrogens. Patients currently receiving antidiabetic treatment or insulin-sensitizing agents were also excluded. The Institutional Review Board (IRB) of the Inje University Haeundae Paik Hospital approved this study and waived the requirement for informed consent.
2. Measurement of anthropometric parameters and ultrasound examinations
All study participants underwent an evaluation of clinical parameters, including age, parity, height, body weight, body mass index (BMI), waist circumference, hip circumference, and waist-to-hip ratio, at their initial outpatient visit. Transvaginal or transrectal pelvic ultrasound examinations were performed during the early follicular phase using a Voluson LOGIQ S7 Ultrasound System (GE Ultrasound Korea, Ltd., Seongnam, Korea) with a microconvex intracavitary probe (frequency range: 3.6–9.0 MHz). The criteria for PCOM included ≥20 follicles measuring 2–9 mm in diameter per ovary and/or an ovarian volume >10 mL, according to the 2018 international evidence-based guideline criteria [4]. A single experienced sonographer specializing in reproductive endocrinology performed all examinations to minimize interobserver variability. Although patients were clinically enrolled between 2010 and 2013, all ultrasound images and clinical records were retrospectively reevaluated to confirm that each participant met this updated morphological threshold, ensuring diagnostic consistency with the current standards throughout the cohort.
3. Biochemical measurements and assessment of insulin sensitivity
Following overnight fasting, blood samples were obtained from all participants during the early follicular phase, in accordance with the Declaration of Helsinki guidelines. For spontaneously menstruating patients, ultrasound examinations and fasting blood collection were conducted between days 2 and 5 of the menstrual cycle. For patients with persistent amenorrhea, a urinary pregnancy test was first performed to confirm the negative results. Subsequently, the early follicular phase status was first verified by ultrasonography, confirming that both the endometrial appearance and follicular status were consistent with the early follicular phase and specifically ensuring that the maximum diameter of any follicle did not exceed 9 mm. Subsequently, the fasting blood samples were collected.
TT and DHEA-S levels were determined using the Elecsys® Testosterone II and DHEA-S assays, respectively (Roche, Indianapolis, IN, USA). FT and 17-OHP levels were assessed using the Free Testo-RIA-CT and 17a-OH hydroxyprogesterone- RIA-CT tests, respectively (DIA Source, Louvain, Belgium). Serum glucose concentrations were determined using an L-Type GluI device (Wako, Osaka, Japan), and insulin levels were assessed with the Elecsys® Insulin assay (Roche, Basel, Switzerland). For all analytical methods, both the intra- and inter-assay coefficients of variation were <8%.
Fasting insulin sensitivity assessment indices (ISAIs) derived from fasting insulin and glucose levels, including the fasting insulin level, glucose-to-insulin ratio (GIR), homeostatic model assessment of insulin resistance (HOMA-IR), and quantitative insulin sensitivity check index (QUICKI), were calculated according to the formulas described by Kwon et al. [16]. Participants were categorized as having abnormal insulin sensitivity if they met one or more of the following validated ISAI thresholds for Asian women with PCOS [16]: fasting insulin level ≥15 μIU/mL [17], HOMA-IR ≥2.64 [13], GIR ≤10.7, or QUICKI ≤0.34 [18].
4. Statistical analyses
Data are presented as mean±standard deviation or median values with ranges. Unpaired t-tests were used for between-group comparisons of continuous variables and the nonparametric Mann-Whitney U-test was used for non-normally distributed (skewed) variables. The relationships between androgen levels (TT, FT, DHEA-S, and 17-OHP levels) and IR parameters were evaluated using Spearman’s rank correlation coefficients. Partial correlation analysis was conducted to assess whether the associations between androgen levels and IR parameters persisted after adjusting for age and BMI, as these anthropometric variables are known confounders of the androgen-insulin relationship. ROC curve methodology was applied to establish optimal androgen cutoff values for detecting abnormal insulin sensitivity. The optimal threshold was determined using Youden’s index, which maximizes the combined sensitivity and specificity. The discriminatory performance of each androgen was assessed using the area under the ROC curve (AUC), with values of 0.7–0.9 indicating moderately accurate and values of 0.5–0.7 suggesting less accurate predictions [19,20]. Statistical analyses were performed using IBM SPSS Statistics version 25.0 (IBM Corp., Armonk, NY, USA). A P-value <0.05 was considered statistically significant.
Results
This study included 180 women with PCOS. The mean age of the participants was 25.92±5.13 years, and the mean BMI was 22.59±5.56 kg/m2. Using the established ISAI criteria, the participants were categorized as having normal or abnormal insulin sensitivity. Of 180 participants, 91 (50.6%) had abnormal insulin sensitivity. Table 1 presents the baseline clinical and biochemical characteristics of the participants.
Baseline clinical and biochemical characteristics of study participants
Spearman’s correlation analysis revealed that TT, FT, and DHEA-S were significantly associated with multiple IR indices, whereas 17-OHP showed no significant correlation (Table 2). FT exhibited the strongest correlations with IR parameters: fasting insulin level (r=0.497; P<0.001), HOMA-IR (r=0.494; P<0.001), GIR (r=−0.492; P<0.001), and QUICKI (r=−0.495; P<0.001). TT and DHEA-S also demonstrated significant correlations with these parameters, although the correlation coefficients were lower than those observed for FT. Partial correlation analysis adjusted for age and BMI revealed that the associations between FT and IR parameters remained statistically significant (data not shown).
Correlations between a variety of androgen levels and IR-related clinical and biochemical parameters in women with polycystic ovary syndrome
ROC curve analyses indicated that FT had the highest predictive accuracy for detecting impaired insulin sensitivity (AUC, 0.759; 95% confidence interval [CI], 0.687–0.831; P<0.001), compared with TT (AUC, 0.674), DHEA-S (AUC, 0.613), and 17-OHP (AUC, 0.496) (Table 3, Fig. 1). In ROC curve analysis, only FT achieved moderately accurate predictive performance, while TT, DHEA-S, and 17-OHP provided less accurate predictions. Based on Youden’s index, the optimal FT cutoff for identifying impaired insulin sensitivity was 1.325 pg/mL (sensitivity, 77.8%; specificity, 64.0%) (Fig. 2).
Area under the receiver operating characteristic curves of androgens for predicting abnormal insulin sensitivity in women with polycystic ovary syndrome
Receiver operating characteristic curve analysis of the predictive accuracy of various androgens for insulin sensitivity in women with PCOS. PCOS, polycystic ovary syndrome.
Receiver operating characteristic curve analysis of the predictive accuracy of FT levels for impaired insulin sensitivity in women with PCOS. AUC, area under the curve; FT, free testosterone; PCOS, polycystic ovary syndrome.
When participants were stratified according to this FT cutoff, those with higher FT levels (≥1.325 pg/mL) exhibited significantly greater IR than those with levels below the threshold. The fasting insulin level, HOMA-IR, and fasting glucose levels were higher and GIR and QUICKI scores were lower in the high-FT group (group 2), consistent with reduced insulin sensitivity (Table 4).
Insulin resistance profiles stratified by free testosterone cutoff in women with polycystic ovary syndrome
Discussion
IR and compensatory hyperinsulinemia play central pathophysiological roles in PCOS. However, in routine clinical practice, i.e., outside of tertiary care settings, such as university hospitals, insulin assays are not readily available in many clinical environments. Furthermore, recent international PCOS guidelines state that “insulin resistance is a pathophysiological factor in PCOS; however, clinically available insulin assays are of limited clinical relevance and are not recommended in routine care” [21]. Since current insulin assays lack sufficient clinical utility and are excluded from routine clinical recommendations, it has become necessary to identify complementary opportunistic indicators of IR among the mandatory evaluative markers for PCOS. Consequently, another core aim of this study was to assess the potential of androgens, essential elements in diagnosing PCOS, as opportunistic metabolic indicators.
Among the various androgens, FT showed the strongest association with IR indices and the highest predictive performance for identifying abnormal insulin sensitivity in women with PCOS. A serum FT concentration of 1.325 pg/mL was determined to be the optimal cutoff value, yielding moderate predictive accuracy (AUC, 0.759; sensitivity, 77.8%; specificity, 64.0%). Women with FT levels above this threshold exhibited significantly higher fasting insulin and HOMA-IR values as well as lower GIR and QUICKI values, indicating impaired insulin sensitivity. These findings suggest that elevated FT may serve as a clinically relevant biochemical marker for identifying women with PCOS who are at increased metabolic risk [8,9,13].
Our findings support the concept that HA and IR in PCOS are bidirectionally related, rather than existing in a simple cause-and-effect relationship [8,9,11]. Our previous work demonstrated that fasting insulin levels ≥9.85 μU/mL could predict the presence of HA (AUC, 0.817) [14], suggesting that insulin dysregulation contributes to androgen excess. The present analysis explored the reverse direction of this relationship, demonstrating that androgen excess, specifically elevated FT levels, was associated with impaired insulin sensitivity at a moderate level of discriminatory accuracy (AUC, 0.759). Together, these observations provide empirical support for the bidirectional nature of the androgen-insulin axis in PCOS pathophysiology [8,9], although longitudinal studies are needed to establish temporal causality.
Several biological mechanisms may underlie the association between excess androgens and IR. Experimental studies have shown that androgens can directly impair insulin signaling pathways by reducing insulin receptor substrate-1 phosphorylation and GLUT-4 translocation in the skeletal muscle and adipose tissue, leading to decreased glucose uptake and peripheral insulin sensitivity [9,10]. Excess androgens also promote visceral adiposity and chronic low-grade inflammation, both of which exacerbate IR [11]. Conversely, hyperinsulinemia stimulates androgen production in ovarian theca cells and suppresses the hepatic synthesis of SHBG, thereby increasing circulating FT levels [6,7]. These reciprocal mechanisms support a self-perpetuating metabolic feedback loop, wherein excess androgen and insulin dysregulation reinforce each other in PCOS [8,9].
While previous studies have explored various surrogate markers of IR in PCOS, including SHBG levels, the free androgen index (FAI), and lipid-derived indices [13,16,22,23], few have established clinically actionable thresholds based on direct androgen measurements. For instance, Chen et al. [22] demonstrated that the SHBG level was inversely correlated with the HOMA-IR and identified a cutoff value of 37 nmol/L for predicting IR (AUC, 0.885; sensitivity, 92.9%; specificity, 74.4%). Zhang et al. [23] reported that a higher FAI was independently associated with glucose intolerance in women with PCOS (adjusted odds ratio, 2.48; 95% CI, 1.39–4.43), after adjustment for multiple metabolic and clinical confounders. Kwon et al. [16] proposed the triglyceride-glucose index as a lipid-based surrogate with an optimal threshold of 8.126 (AUC, 0.791; sensitivity, 80.7%; specificity, 68.3%). Collectively, these findings underscore the metabolic heterogeneity of PCOS and highlight the need for direct and clinically interpretable biochemical thresholds. The present study addressed this gap by defining a specific FT cutoff with validated sensitivity and specificity, providing a directly measurable androgen-based threshold for metabolic risk stratification that complements prior SHBG-, FAI-, and lipid-derived approaches while relying on a routinely measured androgen biomarker in standard PCOS evaluation.
These findings have meaningful clinical implications, particularly for opportunistic screening strategies. We do not propose FT as a substitute for fasting insulin levels or HOMA-IR, both of which remain reference standards for IR assessment. As serum androgens, including FT, are routinely assessed as part of the initial PCOS diagnostic workup, the identified threshold of 1.325 pg/mL can be applied to an already-obtained value, allowing clinicians to simultaneously flag women at an elevated metabolic risk without requiring additional tests. This approach is especially relevant for lean women or those without conventional metabolic risk factors, in whom IR is frequently under-recognized and in whom fasting glucose or HOMA-IR values may not yet exceed established diagnostic thresholds. In such cases, an FT level at or above this cutoff should prompt targeted metabolic follow-up, including fasting insulin measurement or oral glucose tolerance testing. Early identification of these at-risk individuals enables timely implementation of preventive strategies, such as lifestyle modification and insulin-sensitizing agents, such as metformin, which have demonstrated benefits for both metabolic and reproductive outcomes in PCOS [24]. The use of a single, readily available biomarker may enhance the cost-effectiveness compared to complex metabolic assessments, particularly in resource-limited settings. Furthermore, longitudinal monitoring of FT levels may serve as a convenient means of tracking metabolic trajectories over time; however, this application requires validation in prospective studies.
This study had several limitations. First, our cross-sectional design cannot infer causation; longitudinal studies with repeated assessments are required to elucidate the temporal dynamics between HA and IR. Second, insulin sensitivity was estimated using fasting-state surrogate indices rather than gold-standard dynamic tests such as the euglycemic-hyperinsulinemic clamp [25], which may not capture postprandial insulin dynamics. Third, the study population consisted exclusively of Korean women, and ethnic variations in androgen metabolism and insulin sensitivity may limit the generalizability [26]. Fourth, residual confounders from unmeasured variables such as dietary patterns, physical activity, family history, and genetic polymorphisms cannot be excluded. Fifth, the proposed FT cutoff of 1.325 pg/mL was derived using a specific immunoassay platform (Free Testo-RIA-CT [DIA Source]) in a Korean clinical setting and may not be directly transferable to other assay systems or populations without method-specific validation. Finally, the proposed FT cutoff value requires replication in independent and ethnically diverse cohorts before clinical adoption.
In summary, this study provides empirical evidence that the androgen-insulin relationship in PCOS is bidirectional [8,9] and quantifiable using routine biochemical markers. Among the androgens evaluated, FT showed the strongest association with the IR parameters and achieved moderate predictive accuracy (AUC, 0.759) in detecting impaired insulin sensitivity in women with PCOS. A serum FT cutoff of 1.325 pg/mL may serve as a clinically informative reference value for metabolic risk stratification, as women exceeding this threshold exhibited significantly impaired insulin sensitivity across multiple validated indices. However, external validation in independent cohorts is warranted before broader clinical adoption. These findings, together with our previous observation that fasting insulin levels predict HA [14], support a bidirectional relationship between androgen excess and IR in PCOS [8,9]. As FT is routinely measured during PCOS diagnostic evaluation, this threshold can be readily applied to identify metabolically at-risk women who may benefit from early intervention, even in the absence of obesity or other conventional risk factors.
Notes
Conflict of interest
The authors declare no conflict of interest.
Ethical approval
This study was conducted in accordance with the guidelines of the Declaration of Helsinki.
Patient consent
This study was approved by the IRB of a Inje University Haeundae Paik Hospital, which waived the requirement for written consent from subjects in the present study.
Funding information
This research received no external funding.