Serum anti-Müllerian hormone: A potential biomarker for polycystic ovary syndrome

Polycystic ovary syndrome (PCOS) is the most common reproductive endocrine disorder in women of reproductive age with a prevalence between four and 22.5 per cent depending on the population studied and the diagnostic criteria used1,2. PCOS is a complex reproductive syndrome characterized by hyperandrogenism (clinical and/or biochemical), ovulatory dysfunction (oligo- and/or anovulation) and polycystic ovary morphology (PCOM) (polycystic and/or enlarged)3. Currently, Rotterdam criteria with phenotypic classification is followed to diagnose PCOS4,5. Hyperandrogenaemia, a cardinal criteria of PCOS, is rarely seen in southeast Asians6-8. The Endocrine Society advises mass spectrometry (MS) based assay for testosterone estimation due to less sensitivity and specificity of direct immunoassay methods9. However, MS-based assay is not readily available and requires a high-end laboratory setup. Hence, there is a need to find a better marker.

Anti-Müllerian hormone (AMH) is a glycoprotein and a member of the transforming growth factor family. It is produced by the granulosa cells of developing ovarian follicles, mainly preantral and small antral follicles10. AMH level does not alter much during phases of a menstrual cycle or with exogenous oestrogen which is an important advantage of AMH as a biomarker11. In the last few years, there is accumulating evidence on the association of AMH and PCOS12,13. Several studies have reported a high level of AMH, in particular with phenotype A and D PCOS14,15, and suggested incorporating it in the diagnosis16-18. There is heterogeneity in AMH levels in different PCOS phenotypes with higher levels in phenotypes A and D15. However, the cut-off value of AMH for PCOS diagnosis was variable across studies18,19. Some investigators reported a level of 4.7 ng/ml or more serum AMH as a strong predictor of PCOS20,21, whereas others reported >5 ng/ml18. Ethnic diversity is another factor associated with variations in AMH levels; generally, Asian women have lower AMH than Caucasians22.

Prior experience on this subject suggests poor association of PCOS and androgens in addition to other biomarkers6-8. Therefore, it is important to investigate PCOS cases to find a potential diagnostic biomarker. The primary objective of this study was to evaluate the accuracy, diagnostic cut-off value and power of prediction of AMH in north Indian women with PCOS.

Material & Methods

This study was conducted in the department of Reproductive Biology, All India Institute of Medical Sciences, New Delhi, India, from January 2016 to December 2019 after obtaining approval by the Institutional Ethics Committee. Informed consent was obtained from all the participants included in the study.

All suspected PCOS cases underwent clinical and basic investigations before inclusion in the study. Women with PCOS cases were selected for the study after the evaluation of reproductive and menstrual history, hirsutism, testosterone level and targeted ovarian ultrasonography. A menstrual cycle length of 42 days/more (instead of 35 days) was considered as oligomenorrhoea and at least 182 days for amenorrhoea. Hyperandrogenism was defined as Ferriman–Gallwey (FG) score ≥9 (hirsutism/clinical hyperandrogenism)23 and/or testosterone levels >0.6 ng/ml (laboratory cutoff level/biochemical hyperandrogenism). PCOM was considered when targeted ovarian ultrasonography showed follicle size of 2-9 mm and count was ≥12 in one or both ovaries with ovarian size 10 ml or more (one/both ovaries).

Inclusion criteria: All cases were also evaluated for cortisol, prolactin, TSH, oestradiol (E2), progesterone (P4), FSH, luteinizing hormone (LH), dehydroepiandrosterone sulphate (DHEAS), testosterone, etc. (using Abbott Inc. autoanalyzer; IL, USA) to exclude secondary causes and/or for the diagnosis of PCOS. The 17-hydroxyprogesterone along with other androgens (dihydrotestosterone and androstenedione) and sex hormone-binding globulin (SHBG) was measured using commercial enzyme-linked immunosorbent assay (ELISA) kits24. In some cases, chromosome analysis was also carried out to exclude rare secondary causes such as the disorder of sex development/sex reversal. After preliminary evaluation, 131 cases were found as primary PCOS and included in the study (after exclusion of secondary causes such as congenital adrenal hyperplasia (CAH), premature ovarian failure, hyperprolactinaemia, androgen-producing tumours and the disorder of sex development).

Study participants were segregated into four groups such as phenotype A (hyperandrogenism, ovulatory dysfunction and PCOM), phenotype B (hyperandrogenism and ovulatory dysfunction), phenotype C (hyperandrogenism and PCOM) and phenotype D (ovulatory dysfunction and PCOM). 49 women with normal menstrual cycle and fertility were included as controls for the comparison. About 50 per cent of control samples were collected from PHC near Ranchi from normal fertile (at least 2 children, last childbirth >2 yr ago; some lactating but resumed menstruation) tribal women <30 yr.

Exclusion criteria: Before the inclusion as PCOS, every case was evaluated for ovarian neoplasm, CAH, adrenal neoplasm, Cushing’s syndrome, thecoma, drug-induced hyperandrogenism, hypothyroidism, hyperprolactinaemia or premature ovarian failure and excluded from the study.

Sample size: Sample size for the study was computed for comparing testosterone level between PCOS and control based on the following assumptions: mean [standard deviation (SD)] in PCOS as 1.5 (0.5) nmol/l and in controls as 1 (0.5) nmol/l with 95 per cent confidence level, 90 per cent power and control: PCOS as 1:2 allocation ratio24. The minimum sample size required was 44 controls and 88 cases.

Statistical analysis: Data were recorded on a pre designed proforma and managed on an Excel spreadsheet. All the entries were checked for any possible keyboard error. Quantitative variables were assessed for approximate normality. Variables following approximate normal distribution were summarized by mean and SD. Non-normal variables were summarized by a median with the minimum and maximum values. Student’s t test was used to compare the mean between two independent groups, while Wilcoxon rank-sum test was used to compare the median between the two groups. Spearman’s rank correlation was used to compute the correlation coefficient. The receiver operating characteristic (ROC) curve was used to determine the appropriate cut-off value and its predictive value for each of the potential biomarkers of PCOS (total PCOS and phenotype A PCOS cases). All the potential biomarkers were simultaneously considered in a stepwise multivariable binary logistic regression model to determine the weightage/score for predicting PCOS. Using ROC analysis, a cut-off value of the score was determined for the combination of biomarkers for predicting PCOS. Sensitivity, specificity and its 95 per cent confidence interval were computed. In this study, a P value less than 0.05 were considered significant. STATA 15.0 software (Stata Corp. TX, USA) was used for data analysis.

Results

A total of 131 women with PCOS and 49 healthy women were enrolled for this study. Of the 131 PCOS women, 87 were phenotype A, 23 were phenotype B, 8 were phenotype C, and 13 were phenotype D. The mean age of the PCOS group was 23.5 yr and 26.2 yr in control group; the difference between groups being significant (P<0.001). The mean age was lower in the PCOS group, more so with phenotype D (Table I).

High BMI (≥25 kg/m²) was observed in about 50 per cent of PCOS cases. Clinical hyperandrogenism, i.e., FG score of 9 or more, was observed in overall 74 per cent (97/131) total PCOS cases and 77 per cent (67/87) with phenotype A PCOS cases (Table I). Ultrasonographic PCOM of type 1 (ovarian volume 11-15 ml and /or follicle count 12-15 of 2-9 mm size in one/both ovaries) was observed in most (about 71%) of the PCOS cases, whereas type 3 PCOM (ovarian volume >20 ml and/or follicle count >20 of 2-9 mm size in one/both ovary) was seen in about seven per cent of total PCOS cases. Table I summarizes each biomarker analysed. As compared to controls all biomarkers in PCOS group were significantly different.

Table II shows AMH level comparison between age groups (<20 vs. ≥20 yr), BMI groups (<25 vs. ≥25 kg/m²), and PCOM types (type 1/ ovarian volume 11-15 ml and /or follicle count 12-15 of 2-9 mm size in one/both ovaries vs. type 2/ovarian volume 16-20 ml and/or follicle count 16-20 of 2-9 mm size in one/both ovaries). No significant differences were found in AMH levels across age groups, BMI groups, or PCOM types. Spearman’s correlation study of AMH concentration with clinical and biochemical variables of total PCOS and phenotype A PCOS cases found no significant correlations except with androgens (Table III). ROC curve analysis (Table IV) was performed for AMH, T, FAI, DHT, androstenedione, DHEAS, and LH in total PCOS and phenotype A PCOS cases. The area under the ROC curve (AUC) of >0.9 (promising biomarker) was observed with AMH and testosterone. The prediction model for PCOS diagnosis using multivariable binary logistic regression analysis of the biomarkers studied are shown in Table V and Figure. Phenotypes B, C, and D were excluded from analysis due to the small number of cases.

Close

Figure

ROC curve for PCOS prediction model. Stepwise multivariable binary logistic regression model used to predict power of each biomarker. Cut-off score for prediction of PCOS: ≥39.56 with sensitivity (95% CI) of 91% (90.7, 96.8) and specificity (95% CI) 89.9% (88.5, 95.3). ROC, receiver operating characteristics.

Export to PPT

Anti- Müllerian hormone (AMH): The median serum levels of AMH was 8.5 ng/ml in the PCOS group and 2.4 ng/ml in the control group (Table I). Phenotype wise median AMH values were 8.4 ng/ml in phenotype A, 8.9 mg/ml in phenotype B, 5.6 ng/ml in phenotype C and 10.2 ng/ml in phenotype D. Significant difference was found between controls and total (P<0.001) and phenotype A (P<0.05) PCOS cases, respectively. A weak correlation with AMH levels was observed with testosterone in total PCOS cases and DHT, as well as age in phenotype A PCOS cases. However, we found no correlation between AMH level with BMI, FG score, and PCOM (Table III). The area under the ROC curve (AUC) for the AMH assay yielded a highly satisfactory result. The AUC for AMH was 0.93 [95% confidence interval (CI): 0.88, 0.96] in total PCOS cases and 0.96 (95% CI: 0.91, 0.98) for phenotype A PCOS cases. A high level of AMH was observed in 84 per cent of the total (cut-off level >4.1 ng/ml) and 87.4 per cent of the phenotype A (cut-off value 5.17 ng/ml) PCOS cases (Table IV). The prediction model for PCOS using multivariable binary logistic regression detects AMH as the best marker (P<0.001) with the highest weighted score of 27.3 and a cut-off score for prediction of PCOS as ≥39.51 (Table V; Figure).

Testosterone: The mean testosterone was 0.5 ng/ml in the study group, whereas in the control group, it was 0.23 ng/ml (Table I). A significant (P<0.05) difference was observed between the controls with total and phenotype A PCOS cases (phenotypes B, C, and D were excluded from analysis due to the small number of cases). A weak correlation between testosterone and AMH was also observed in total PCOS cases. The AUC for testosterone yielded satisfactory results in total (0.87; 95% CI: 0.81, 0.91) and phenotype A (0.92; 95% CI: 0.87, 0.96) PCOS cases (Table IV). A high testosterone level was observed in 80.8 per cent of the total PCOS cases with a cut-off value of 0.3 ng/ml and in 83.7 per cent of the phenotype A PCOS cases with a cut-off value of 0.35 ng/ml (Table IV). The prediction model for PCOS using multivariable binary logistic regression detects testosterone as the second-best (next to AMH) marker (P<0.001) with the highest weighted score of 24.3 (Table V; Figure).

Dihydrotestosterone (DHT): The mean DHT value was 561 pg/ml in the study group, whereas, it was significantly low in the control group at 257 pg/ml. A weak correlation between DHT and AMH was observed in phenotype A PCOS cases. The AUC for DHT yielded satisfactory results in total (0.87; 95% CI: 0.81, 0.92) and phenotype A (0.91; 95% CI: 0.84, 0.95) PCOS cases (Table IV). A high level of DHT was observed in about 80 per cent of PCOS cases with a cut-off value of 360 pg/ml for total and 370 pg/ml for phenotype A PCOS cases (Table IV). The prediction model for PCOS using multivariable binary logistic regression detects DHT as a good marker (P<0.007) with the highest weighted score of 11.2 and a cut-off score for the prediction of PCOS as ≥39.51 (Table V; Figure).

Androstenedione: The median androstenedione was 1.1 ng/ml in the study group, whereas in the control group, it was 0.9 ng/ml (P<0.05). No correlation between AMH level and androstenedione was observed. The AUC for androstenedione was 0.77 in total and 0.74 in phenotype A PCOS cases. A high level of androstenedione was observed in about 62 per cent of PCOS cases, with a cut-off value of 1 ng/ml for total and phenotype A PCOS cases (Table IV). The prediction model for PCOS using multivariable binary logistic regression detects androstenedione as a marker with the highest weighted score of 12.2 and a cut-off score for prediction of PCOS as ≥39.51 (Table V; Figure).

Dehydroepiandrosterone sulphate (DHEAS): The mean DHEAS was 216 μg/dl in the study group, whereas in the control group, it was 152 μg/dl. No correlation between AMH level and DHEAS was observed. The AUC for DHEAS was 0.72 in total and 0.7 in phenotype A PCOS cases. A high level of DHEAS was observed in about 62 per cent of total PCOS cases with a cut-off value of 164 μg/dl and in about 58 per cent of phenotype A PCOS cases with a cut-off value of 166 μg/dl (Table IV). The prediction model for PCOS using multivariable binary logistic regression detects androstenedione as a marker (P<0.018) with the highest weighted score of 10 and a cut-off score for prediction of PCOS as ≥39.51 (Table V; Figure).

Free androgen index (FAI): The median FAI was 3 in the study group, whereas in the control group, it was 0.97 (P<0.001). No correlation between AMH level and FAI was observed. The AUC for FAI was 0.84 in total and 0.86 in phenotype A PCOS cases. A high level of FAI was observed in about 70 per cent of PCOS cases with a cut-off value of 2.2 for total and 2.9 for phenotype A PCOS case (Table IV). The prediction model for PCOS using multivariable binary logistic regression detects no significance with FAI for the prediction of PCOS.

Luteinizing hormone (LH): The median LH was 4.6 mIU/ml in the study group, whereas in the control group, it was 4.2, which is insignificant. The phenotype wise median LH value was 7.3 mIU/ml in phenotype A (P<0.001), 4.7 mIU/ml in phenotype B, 3.5 mIU/ml in phenotype C and 6.3 mIU/ml in phenotype D. No correlation between AMH level and LH was observed. The AUC for LH was 0.72 in total and 0.74 in phenotype A PCOS cases. A high level of LH was observed in about 84 per cent of PCOS cases, with a cut-off value of 4.6 for total and 4.3 for phenotype A PCOS cases (Table I). The prediction model for PCOS using multivariable binary logistic regression detected no significance value for LH as a predictor of PCOS.

Discussion

PCOS is a complex reproductive endocrine disorder characterized by hyperandrogenism (clinical and/or biochemical), chronic ovulatory dysfunction (oligo and/or anovulation), and PCOM (polycystic and/or enlarged ovary). The first scientific description of PCOS was published in 1935 by Stein and Levanthal25. Since then, many efforts have been made for the precise diagnosis; however, unfortunately till date, there is no promising biomarker (sensitive and specific) and no definite etiologic factors detectable in primary PCOS. Serum testosterone and/or FAI is commonly used as hyperandrogenaemia markers26; however, high values are rarely observed in women with PCOS from southeast Asian countries, including India6-8. Similarly, other androgens, such as androstenedione and DHEAS, are weakly correlated with PCOS. There is no other biochemical marker linked to PCOS until recently. In recent years, several studies have reported the diagnostic value of AMH in PCOS patients14-18,27. The most common PCOS phenotype in our study was phenotype A (all three features, i.e. classical PCOS), about 66 per cent and similar to most western countries28. However, there are more PCOS cases with phenotype D6,7.

This study reports the reference range of AMH for diagnosing PCOS in the north Indian population. It also compares the diagnostic value of each biomarker of PCOS, such as testosterone, FAI, DHT, androstenedione, DHEAS, and LH. In this study, we have found a median AMH value of 2.4 ng/ml in controls whereas about 8.5 ng/ml in PCOS, i.e., nearly 3.5 times more indicating a substantial diagnostic value (P<0.001). The maximum AMH median value was observed in phenotype D (10.2 ng/ml), followed by phenotype B (8.9 ng/ml), and least with phenotype C (5.6 ng/ml). We have also compared the association of AMH with BMI (<25 and ≥25 kg/m²), FG score (<9 and ≥9), as well as PCOM but found no relationship. Failure to find any significant correlation between PCOM and AMH could be due to the subjective nature of PCOM by abdominal ultrasonography (both operator and patient, particularly with obesity; ~50 per cent of PCOS cases were overweight). Similarly, Spearman’s correlation of AMH (ng/ml) of total and phenotype A PCOS cases with biochemical variables such as FAI, androstenedione, DHEAS, and LH failed to detect any association.

The AMH cut-off value was derived from ROC curve analysis as 4.1 ng/ml in total PCOS cases (5.17 ng/ml for phenotype A). These cut-off values picked up 84 percent of total and 87 per cent of phenotype A PCOS cases. This association was much better than any other biomarkers and the FG score of ≥9 was observed in about 74 per cent of PCOS cases. The association of PCOS with other biomarkers was 81 per cent with testosterone, 78 per cent with DHT, 62 per cent with androstenedione, 70 per cent with FAI, 62 per cent with DHEAS, etc., using respective ROC-derived cut-off values. This finding showed that all women with PCOS, even those who did not have PCOM (phenotype B) or hyperandrogenism (phenotype D), had significantly higher AMH levels, indicating that AMH is associated with PCOS more strongly than any other biomarker of PCOS, irrespective of PCOS phenotypes. Various cut-off values of AMH have been reported in the literature and vary from 4.7 ng/ml20,21 to 5 ng/ml18,22,29 in Caucasians or even 10 ng/ml30,31 in Japanese and Korean women. Our cut-off values (4.1 ng/ml for total and 5.17 ng/ml for phenotype A PCOS cases) are similar to Caucasians,18,20-22,29 and much lower than orientals30,31. Phenotypes B, C and D were excluded from analyses due to the small number of cases. Serum AMH, as a diagnostic marker of PCOS was found to have high sensitivity (84%) and specificity (88%). ROC curve analysis suggests that AMH is superior to other biomarkers studied here in diagnosing PCOS. The area under the curve (AUC) of ROC was found to be 0.93 (95% CI: 0.88, 0.96) for total PCOS cases and 0.96 (95% CI: 0.91, 0.98) for phenotype A PCOS cases. The AUC of ROC of total and phenotype A PCOS cases was 0.87 and 0.92 with testosterone, 0.87 and 0.91 with DHT, 0.84 and 0.86 with FAI, 0.77 and 0.74 with androstenedione, 0.72 and 0.7 with DHEAS and 0.72 and 0.74 with LH, respectively. An AUC of ROC of more than 0.9 can be considered a strong correlation and may be used as a diagnostic marker. This signifies that AMH is likely to be associated with >93 per cent of total PCOS cases and >96 per cent with phenotype A, considering 4.1 ng/ml (total PCOS cases) or 5.17 ng/ml (phenotype A PCOS cases) as cut-off values. The literature reported sensitivity between 49 per cent and 74 per cent when the specificity was set at 92 per cent; however, others reported higher sensitivity with a little lower specificity16,28,31,32 similar to the present study (specificity kept at 88 and 90%). The AMH estimation seems comparatively simple, sensitive, and the best available marker associated with PCOS.

In this study, the mean age of PCOS cases was significantly less than controls, in particular with phenotype D. Relatively younger age in PCOS was also reported by other studies33,34. Similarly, the mean BMI in PCOS cases (25.4 kg/m²) was significantly higher than controls (22.6 kg/m²; P<0.01) as in other reports35. High BMI was observed with phenotypes A (26.1 kg/m²) and B (25.3 kg/m²), whereas phenotypes C (22.9 kg/m²) and D (22.3 kg/m²) were normal. High BMI was observed in 49 per cent of PCOS cases and 22 per cent in controls. In this study, unlike other studies36,37, LH level in the total PCOS group (median: 4.6 mIU/ml) was similar to the control group (median: 4.2 mIU/ml). However, a significant difference was observed with phenotype A (median: 7.3 mIU/ml). Spearman’s correlation of AMH concentration with clinical and biochemical variables of total and phenotype A PCOS cases did not find any association with BMI, FG score, PCOM, FAI, androstenedione, DHEAS, and LH; however, a weak correlation was associated with androgens (testosterone in total and DHT in phenotype A PCOS cases). Other studies38,39 also have shown some association of AMH with androgens in PCOS. The prediction model for PCOS using multivariable binary logistic regression detects AMH as the best marker associated with PCOS, followed by testosterone and DHT. AMH could play an essential role in the diagnosis of PCOS, in particular cases with normal androgen levels, commonly observed in southeast Asian countries. Moreover, accurate estimation of testosterone is difficult. The incorporation of AMH will also help to differentiate PCOS like condition associated with non-classical CAH and premature ovarian insufficiency/failure.

In this study, serum testosterone was found to be the second-best biomarker of PCOS. However, the cut-off value of testosterone (0.3 ng/ml) was found to be much lower than that reported in literature (0.6 ng/ml)40,41. This could be true as high testosterone is rarely observed in Asian women with PCOS6,7, thus it is time to think on race-specific cut-off value.

The major strength is a phenotype-wise characterized homogeneous group of primary PCOS cases. All secondary PCOS cases (CAH, gonadal tumour, premature ovarian failure, the disorder of sex development, sex reversal, etc.) were excluded by appropriate investigations. The weakness of this study is the fewer cases of phenotypes B, C, and D besides the relatively younger age group (23.5 yr in the study group and 26.2 yr in controls).

Overall, our results suggest that AMH is the best available biomarker with a maximum AUC of ROC (>0.9) and prediction score (weight/point) and has the potential as the diagnostic biomarker of PCOS. The cut-off level of AMH for diagnosing PCOS was 4.1 ng/ml in total PCOS cases (5.17 ng/ml for phenotype A PCOS cases) from north India and may be implemented in routine hospital laboratories to diagnose PCOS. In addition, AMH is an independent biomarker except for a weak correlation with androgens (testosterone and DHT) in women with PCOS. However, more studies with a large cohort are needed for the validation of this result, in particular with phenotypes B, C, and D.