Adjusting antimüllerian hormone levels for age and body mass index improves detection of polycystic ovary syndrome

Abstract

Objective:

To examine whether accounting for a woman’s age and body mass index (BMI) would improve the ability of antimüllerian hormone (AMH) to distinguish between women with (cases) and without (controls) polycystic ovarian syndrome (PCOS).

Design:

An opportunistic case-control dataset of reproductive age women having evaluations for PCOS as defined by National Institutes of Health criteria.

Setting:

Two medical centers in the United States enrolled women. Serum samples were analyzed for relevant analytes.

Patients:

Women were between 18 and 39 years of age when samples and clinical information were collected. Residual samples had been stored for 2–17 years. AMH was measured via immunoassay.

Interventions:

None; this was an observational study.

Main outcome measures:

Detection and false-positive rates for PCOS were computed for AMH results expressed as multiples of the median (MoM) both before and after adjustment for the woman’s age and BMI.

Results:

Using unadjusted AMH MoM results, 168 cases (78%) cases were at or beyond the 90^th centile of controls (2.47 MoM). After accounting for each woman’s age and BMI, 188 (87%) of those women were beyond the 90^th centile of controls (2.20 MoM), a significant increase (P = .015). The adjusted AMH MoM levels fitted logarithmic normal distributions well (mean, standard deviation for controls and cases of 0.0000, 0.2765 and 0.6884, 0.2874, respectively) and this allowed for computation of patient-specific PCOS risks.

Conclusions:

Accounting for the woman’s age and BMI resulted in significantly higher AMH-based detection rates for PCOS at a 10% false-positive rate, and patient-specific PCOS risks could be computed.

Polycystic ovary syndrome (PCOS) is a common endocrine disorder affecting between 8% and 13% of reproductive-aged women, depending on the defining characteristics and population studied (1–3). Prevalence using the 1990 National Institutes of Health (NIH) criteria (4) appears to be similar in countries such as the United States, the United Kingdom, Spain, Greece, and Australia (5). There is some evidence that the prevalence in the United States may vary by region, with higher rates in the southern United States and lower rates in the northeastern United States (5). There is growing interest in the role and measurement of AMH in the identification of PCOS (6–8). PCOS is a syndrome characterized by hyperandrogenism, ovulatory dysfunction, and polycystic ovary morphology (7). Because the assessment of ovarian morphology requires ultrasonography, there has been considerable interest in identifying biochemical markers for PCOS-associated changes in folliculogenesis (7). Antimüllerian hormone (AMH), also known as Müllerian inhibiting substance, may be such a biomarker (9), with higher AMH levels associated with the presence of PCOS.

AMH is a member of the transforming growth factor – β (TGFβ) family secreted as a 140-kDa full-length homodimeric precursor and then cleaved by prohormone convertases to generate a 110-kDa pro N-terminal and a 25-kDa mature C-terminal dimers. The cleaved C-terminal region of AMH remains noncovalently associated with the cleaved N-terminal region (10). Further proteolytic cleavage at dibasic or monobasic sites located between the two domains of the noncovalent AMH complex results in dissociation of the N-terminal region (11). Recent findings demonstrate that the processing of AMH is tightly regulated. The mature C-terminal isoforms of AMH (12.5-kDa and 16-kDa) are present in the nucleus, whereas the pro-mature form is expressed in the cytoplasm of human granulosa cells from small antral follicles (12, 13). The pro-mature noncovalently associated complex of AMH is the predominant form found in follicular fluid and in the circulation, and it is the biologically active form of AMH. Serum AMH concentrations in women gradually decline throughout reproductive life, becoming undetectable by menopause. Reports on the effects of obesity on serum AMH levels have been conflicting. However, a recent study has shown that elevated body mass index (BMI) negatively correlates with AMH in women of white ethnicity (14). Within the ovary, highest levels of AMH are expressed by the granulosa cells of small antral follicles <4 mm in diameter (14). Serum AMH concentrations correlate strongly with the number of early antral follicle counts determined by transvaginal ultrasound. A high circulating AMH concentration identifies women with an unusually high number of small antral follicles. Replacing polycystic ovary morphology with serum AMH concentrations has been reported to identify PCOS with a high specificity and sensitivity (15). Specificity and sensitivity are dependent on the AMH concentration cut-offs used for optimal detection of women affected by PCOS. Given that AMH concentrations are relative to age and obesity, one cannot apply a universal serum AMH cut-off level for women of all ages and all sizes for the detection of PCOS.

Our hypothesis was that by adjusting serum AMH levels for age and BMI, the detection rate for PCOS (clinical sensitivity), at a given false-positive rate, would increase significantly.

MATERIALS AND METHODS

Identification of Women With and Without PCOS

This study used an existing opportunistic case/control dataset. Participants were healthy women of white ethnicity and European ancestry studied between 1998 and 2013, with ages between 18 and 40 years. None of the women were taking medications affecting gonadal function or glucose metabolism for at least 1 month prior to the study. The exception was for contraceptive steroids, which were stopped at least 3 months prior to the study. PCOS was diagnosed in 215 women by the NIH criteria of hyperandrogenism and eight or fewer menses per year, with exclusion of other hyperandrogenic disorders of the ovaries, adrenals, and pituitary (4). All women with PCOS (cases) had elevated total and/or bioavailable testosterone levels measured at the initial screening visit (16). A total of 111 control women had normal reproductive histories, menses every 27–35 days, no hirsutism (Ferriman-Gallwey score <8) (17), and normal circulating androgen levels. The NIH criteria were chosen because the aims of the original study included studying the PCOS metabolic phenotype.

Ethical Approval

The Institutional Review Boards of the Feinberg School of Medicine, Northwestern University, and Penn State University School of Medicine approved this study. Written informed consent was obtained from all subjects prior to participation. Clinical and biochemical data (excluding AMH) from some of the study subjects have been previously reported (18–20). A small number of women from the previous studies did not have sufficient serum samples available for testing and were excluded.

Collection of Serum Samples

Early-morning fasting blood samples were available for reproductive hormone testing in the 326 women studied. In three controls, the luteinizing hormone (LH), follicle stimulating hormone (FSH) and LH/FSH ratio were not available, and for four cases, the sex hormone binding globulin (SHBG) and free androgen index were not available. All missing data were due to sample exhaustion. Serum samples were stored at −80°C until assayed in January 2015 at Ansh Labs (Webster, TX). Laboratory technicians were blinded to the subject phenotype. Specimens from both controls and PCOS-affected women were collected in the same way, subjected to the same storage conditions, and stored for the same range of years to minimize potential bias.

Serum Assays

Testosterone and DHEAS were measured by radioimmunoassay (DPC, now Siemens, Los Angeles, CA) as previously reported (21). SHBG was measured by immunoradiometric assay (DSL, Webster, TX) in 44 control and 167 PCOS women, and by Immulite (Siemens, Tarrytown, NY) in 67 control and 48 PCOS women (16). In the 44 control and 167 PCOS women in whom SHBG was measured by immunoradiometric assay, bioavailable testosterone was directly measured by ammonium sulfate precipitation (16). Bioavailable testosterone was calculated from testosterone and SHBG concentrations measured by Immulite as reported (22) in 67 control and 48 PCOS-affected women.

Measurement of AMH

A commercially available AMH enzyme-linked immunosorbent assay (ELISA) (Ansh Labs, Webster, TX) was used to test all samples. The assay’s analytical range is 0.003–15.0 ng/mL. Reproducibility of the AMH ELISA was determined by calculating the between-plate coefficient of variation (CV) on low and high controls (1.41 and 4.43 ng/mL). The CVs observed over 22 runs/plates at these concentrations were 6.4% and 6.8%, respectively. The limit of detection of the assay, defined as the lowest concentration of AMH that could be distinguished from the zero calibrator, was 0.002 ng/mL. The calibration range for AMH was 0.003 to 0.750 ng/mL, and all controls and samples were diluted 20-fold prior to measurement.

Statistical Analysis

Analysis was restricted to those women who provided a serum specimen and were classified as being a control (non-PCOS) or case (diagnosis of PCOS). AMH values were converted to MoM levels based on medians derived only from control subjects. The first step is to derive a median (or medians) using AMH data from the control women. Age was originally collected in truncated years, so 0.5 year was added to each so that the equations would be valid for decimal ages as well (e.g., a truncated age of 20 is equivalent to a decimal age of 20.5). When there was no adjustment for age and/or BMI, the median AMH level observed in the 111 control women was used. When age and BMI were accounted for, a regression analysis was performed (e.g., log (AMH) = c1*age+c2*BMI+b), and the median value for each woman would be computed using that fitted log linear equation. Although the median(s) were computed using the 111 controls, they were applied to the 215 women with PCOS as well. A normalizing step was then taken to ensure that the median MoM in the 111 controls was exactly 1.00. This was accomplished by dividing all 326 AMH MoM levels by the median AMH MoM level in control women. For example, the median AMH level in the 111 controls was 2.820 ng/mL. If a case and control woman had AMH levels of 4.200 and 17.100 (approximately the 75^th centile of each group), their corresponding unadjusted MoM levels would be 1.49 and 6.06, respectively.

In the first analysis, AMH results in unadjusted MoM levels were used, whereas in the second analysis, the AMH MoM levels were adjusted for both the woman’s age and BMI. The PCOS detection rate was defined as the proportion of PCOS cases with AMH levels at or above a specified cut-off (90^th centile of controls). The false-positive rate was defined as the proportion of controls (non-PCOS) with levels above that same cut-off (i.e., 10%). Adjusted MoM levels in the two groups were then analyzed via a probability plot to examine how well they fitted a Gaussian distribution after a logarithmic transformation. These fitted parameters were then used to model screening performance of AMH including the detection rates, false-positive rates, and individualized risks. Hypothesis testing done was by t-test for continuous variables and χ² with Yates correction for categorical variables. Confidence intervals were computed using the Wald method. P values were two-sided, at the 0.05 level. Several variables were highly right-skewed, but a logarithmic transformation allowed for usual parametric statistics to be used. Analysis was performed with the BMPD statistical package (Statistical Solutions, Boston, MA) or GraphPad (GraphPad, San Diego, CA), and graphics produced in Prism 7 (GraphPad, San Diego, CA).

RESULTS

Population Description

Baseline characteristics for the 326 women included in this study are shown in Table 1, stratified by PCOS status. The ages of the women were similar, but body mass index (BMI) was significantly higher in the PCOS women (P < .001). Likewise, the androgenic hormones tested (including testosterone, free testosterone, and DHEA-S) were all significantly elevated in PCOS-affected women vs. controls (P < .001). Using a testosterone cut-off level of 45 ng/dL, 5% of control and 88% of women with PCOS had elevations (P < .001, Supplemental Figure S1). SHBG was decreased in PCOS-affected women, yielding higher calculated free androgen index values in PCOS-affected women vs. controls (P < .001).

TABLE 1.

Patient baseline demographics of the study population, stratified by the presence or absence of polycystic ovarian syndrome (PCOS)

Variable	Controls	PCOS	P value
Number of women	111	215
Age, yr
Median (range)	29(18 – 39)	28(18 – 39)
Mean (SD)	29.0 (5.7)	27.9 (4.4)	.057
Body mass index, kg/m2
Median (range)	28.1 (19.0 – 55.6)	35.4(18.8 – 60.3)
Mean (SD)	29.4(7.8)	35.8 (8.7)	< .001
Testosterone, ng/dL
Median (range)	25 (5 – 53)	73 (28 – 197)
Mean (SDa)	27.0 (0.19)	76.8(0.15)	< .001
Free testosterone, ng/dL
Median (range)	6(1 – 15)	25(8 – 81)
Mean (SD)	7.0 (0.23)	27.9(0.19)	< .001
SHBG, nmol/L
Median (range)	66(10 – 292)	48 (9 – 250)
Mean (SDa)	74.1 (0.24)	56.8 (0.26)	< .001
FAIb
Median (range)	1.26 (0.18 – 4.86)	5.31 (1.03 – 40.1)
Mean (SDa)	1.58 (0.26)	6.49 (0.26)	< .001
DHEA-S, ng/mL
Median (range)	1340 (370 – 2650)	2063 (486 – 8252)
Mean (SDa)	1380(0.19)	2220 (0.22)	< .001
Luteinizing hormone, IU/mL
Median (range)	4.0(0.80 – 62.3)	11.0(2.6 – 46.0)
Mean (SDa)	7.4(0.25)	12.4 (0.35)	< .001
Follicle-stimulating hormone, mIU/mL
Median (range)	5.7 (1.5 – 42.2)	8.0 (1.3 – 16.0)
Mean (SDa)	6.9(0.29)	8.0 (0.21)	.002
LH/FSH ratio
Median (range)	0.84 (0.19 – 6.00)	1.50 (0.36 – 7.60)
Mean (SDa)	1.18 (0.031)	1.75 (0.018)	< .001
Antimüllerian hormone, ng/mL
Median (range)	2.82 (0.02 – 11.4)	11.2(0.43 – 51.9)
Mean (SDa)	3.3 (0.43)	12.4 (0.28)	< .001

DHEA-S = dehydroepiandrosterone sulfate; FAI = free androgen index; LH/FSH = ratio of luteinizing hormone to follicle-stimulating hormone; SD = standard deviation; SHBG = sex hormone-binding globulin.

In three controls, the LH, FSH and LH/FSH ratio were not available, and in four cases, SHBG and FAI were not available because of sample exhaustion.

^aLogarithmic standard deviation: P value for comparison of the logarithm of the medians and standard deviations between groups.

^bFree androgen index (FAI) = [Testosterone (nmol/L)] •100)/[SHBG (nmol/L)]; where [T(nmol/L)] = T(ng/dL) • 0.0347, and biochemical hyperandrogenism = FAI >4.5.

Accounting for Age and BMI

Linear regression of the woman’s age alone, and the woman’s age and BMI, against AMH levels (after logarithmic transformation) were performed using data from the 111 control women. Both independent variables were strongly associated with AMH (P < .001). Table 2 shows the expected AMH values for selected ages (column 3) and ages and BMIs (column 4) in control women, along with the regression equations. For example, if only age were accounted for, a 30-year old woman would have an expected AMH of 2.24 ng/mL regardless of her BMI (column 2). If both age and BMI were accounted for, the expected AMH levels would vary from 3.00 ng/mL for a BMI of 20 kg/m² to 1.67 ng/mL for a BMI of 40 kg/m². This demonstrates the impact of including BMI when interpreting AMH levels. All AMH results in mass units were then converted to MoM by dividing them by the expected AMH result using the requisite regression equation. In this manner, the resulting AMH MoM levels are corrected for the woman’s age or her age and BMI. By definition, the median adjusted MoM in the 111 controls will be 1.00. The observed median AMH MoM adjusted for age and BMI in women with PCOS was 4.88.

TABLE 2.

Expected antimüllerian hormone (AMH) median values by age and body mass index (BMI) derived from 111 women without polycystic ovarian syndrome (PCOS) (controls)

		Expected AMH (ng/mL)a
Age (yr)a	BMI (kg/m2)	Age aloneb	Age and BMIc
20.5	20	4.14	4.92 (4.2, 5.6)
20.5	30	4.14	3.67 (2.9, 4.4)
20.5	40	4.14	2.73 (2.0, 3.5)
30.5	20	2.24	3.00 (2.3, 3.7)
30.5	30	2.24	2.24(1.5, 3.0)
30.5	40	2.24	1.67 (0.9, 2.4)
40.5	20	1.21	1.83 (1.1, 2.6)
40.5	30	1.21	1.36 (0.6, 2.1)
40.5	40	1.21	1.02 (0.3, 1.7)

Data in parentheses are confidence intervals.

^aAssumes decimal age. Decimal 20.5 years is equivalent to a truncated 25 years.

^bExpected AMH in controls = 10 ^{1.16476 − 0.0267•Age}.

^cExpected AMH in controls = 10 ^{1.38947 − 0.0215•Age − 0.0128•BMI}.

Screening Performance Before and After Adjustments

Figure 1A shows the AMH results in control and case women both with and without adjustment for age and BMI. The “unadjusted” AMH MoM levels were created by dividing each AMH value by 2.82 ng/mL (the median AMH value in nanograms per milliliter among control women). The adjusted AMH MoM have accounted for the woman’s age and BMI. The control MoM levels (open circles) are evenly scattered around the line of identity (Y = X), regardless of whether they were adjusted for age and BMI. However, among the PCOS cases (small filled circles), the adjusted AMH levels tend to fall above the line of identity, indicating that the adjustment for age and BMI systematically increased the interpretive MoM levels in cases. This is likely due to the important difference in the women’s weight and would be expected to improve the separation between cases and controls. Similar plots when the AMH MoM levels are adjusted for age along are shown in Supplemental Figure S2.

FIGURE 1.

Figure 1A shows an analysis of antimüllerian hormone (AMH) levels in women with and without polycystic ovarian syndrome (PCOS). (A) AMH MoM results are shown, with no adjustment for age and body mass index (BMI) (horizontal logarithmic axis) vs. those with adjustment for both age and BMI (vertical logarithmic axis). Filled symbols indicate results in women diagnosed with PCOS (cases) and open symbols indicate women without PCOS (controls). The solid line indicates Y = X. The horizontal arrow indicates where four datapoints were set to either the Y or X value were set to 0.1 for plotting purposes. Figure 1B shows the same data but in a scatterplot format for each group. The datapoints are randomly dithered left or right within a category to reduce overlap. The horizontal line at 2.197 is the 90^th centile of controls. Overall, 10% of the controls (false-positive rate) and 88% of the cases fall above this level. Figure 1C shows a probability plot for the cases and controls after adjustment for age and BMI. The horizontal axis is the Gaussian centile. The short vertical lines indicate the range over which the slope of the line is computed. That slope is an estimate of the logarithmic standard deviation. Figure 1D shows the resulting overlapping curves given the logarithmic mean and standard deviation generated from Figure 1C and provided in the main text. The dashed Gaussian curve represent the controls, whereas the solid curve represents the PCOS cases. The short vertical lines indicate the truncation limits. Outside of these limits, estimating likelihood ratios and patient-specific risks are likely to be less reliable. The thin vertical line is drawn at 1.5 MoM and is used as an example in the main text.

Figure 1B shows a scatterplot of AMH-adjusted MoM levels separately for control and PCOS cases. The line drawn at 2.197 MoM is the 90^th centile of the control population. Among PCOS cases, 87% (188/215, 95% CI 82%−91%) have adjusted AMH MoM levels above this level. This is a significantly higher rate (P = .015) than that found for AMH MoM levels prior to adjustment, where the 90^th centile of the control population is 2.47 MoM and the corresponding PCOS detection rate is 78% (168/215, 95% CI 72%−83%).

Figure 1C further explores the AMH-adjusted MoM levels in controls and PCOS cases via probability plots. The vertical axis shows the AMH levels on a logarithmic scale, and the horizontal axis is the Gaussian centile (based on the z-score). The slope of the fitted line is the standard deviation. The data fitted a straight line reasonably well between the 20^th and 90^th centile in controls and the 5^th and 95^th centiles in cases. The resulting logarithmic means and standard deviations (controls 0.000, 0.2765 and PCOS cases 0.6884, 0.2874) are shown as smooth curves in Figure 1D.

Fitted Logarithmic Gaussian Curves

The advantage of these fitted Gaussian curves is the ability to compute a likelihood ratio (ratio of the case to control ordinal) for any given AMH MoM. These results can then be used to compute patient-specific risks. In addition, these parameters can provide modeled detection rates for any false-positive rate as well as false-positive rates for any detection rate. Such modeling and individual risk estimates have proved highly useful as part of routine prenatal screening for Down syndrome (23, 24).

Modeled Screening Performance and Patient-Specific PCOS Risks

Table 3 shows the screening performance of AMH when expressed as age- and BMI-adjusted MoM levels. The model can be used to provide women with individualized risk estimates for PCOS given one of two selected prior risk levels: 1:9 (or 10%) representing the background odds (risk) of PCOS, and 1:1 (or 50%) representing the odds (risk) of an individual being evaluated with suspected PCOS. Table 3 provides a range of AMH MoM levels with the modeled PCOS detection rate, corresponding false-positive rate, the individual likelihood ratio, and the two individualized risk estimates. AMH levels below 2.2 MoM (the approximate crossover point in Figure 1D) are associated with reduced risks of PCOS, whereas higher levels are associated with increased risk. At a given PCOS MoM level cut-off, the area of the unaffected curve to the right is the false-positive rate, and the area of the PCOS curve is the detection rate. For example, at an AMH MoM of 1.5, for example, 95% of the PCOS and 26% of the unaffected curve is to the left, indicating the detection and false-positive rate, respectively. To generate the likelihood ratio for a given AMH MoM level, the height of the PCOS curve is divided by the height of the unaffected curve (Figure 1D) using the standard Gaussian distribution equation and the provided log means and log standard deviations. At an AMH MoM of 1.50, for example, the two heights are 0.282 and 1.181, resulting in a likelihood ratio of 0.238 or about a fourfold reduction in risk. The use of truncations limits of 1.0 and 4.0 MoM (Figure 1D) assist in ensuring that reliable odds are generated by excluding modeling in the tails of either of the two distributions.

TABLE 3.

Modeled screening performance of serum antimüllerian hormone (AMH), expressed as adjusted multiples of the median (MoM) levels, to identify polycystic ovarian syndrome (PCOS), with associated individual likelihood ratios and individualized risk estimates for two prior risk levels

AMH	PCOS		Likelihood	Patient-specific PCOS riska
MoM	DR (%)	FPR (%)	ratioa	@ 1:9b	@ 1:1b
1.0	99%	50%	0.055	1%	6%
1.5	96%	26%	0.24	3%	19%
2.0	91%	14%	0.70	7%	41%
2.5	84%	7.5%	1.63	15%	62%
3.0	77%	4.2%	3.26	27%	76%
3.5	69%	2.5%	5.9	40%	85%
4.0	62%	1.5%	9.8	52%	91%

DR = detection rate (proportion of all PCOS with an elevated AMH MoM); FPR = false-positive rate (proportion of all non-PCOS controls with an elevated AMH MoM)

^aLikelihood ratio is the relative increase (ratios>1.0) or decrease (ratios<1.0) in the patient’s prior risk due to her AMH MoM level; patient-specific PCOS risk is the estimated proportion of women with this AMH MoM level that will be diagnosed with PCOS, and is computed by multiplying the patient’s a priori risk times the likelihood ratio.

^bPrior risks assumed to be set to 1:9 (simulating background risk of PCOS at 10%) or 1:1 (simulating risk of 50% for an individual with suspected PCOS).

DISCUSSION

AMH is known to be a useful biomarker of reproductive health relative to antral follicle number and healthy ovarian reserve. Specimens were from an opportunistic cohort of age-matched controls and well-characterized PCOS-affected subjects. Our aim was to assess the clinical validity of serum AMH concentrations for the detection of PCOS before and after accounting for age and BMI. To our knowledge, this is the first study to convert serum AMH measurements into age- and BMI-adjusted MoM as a way to improve clinical validity and utility. Converting the AMH to adjusted MoM levels in the two groups achieved significantly higher detection of PCOS (87% vs. 73%) at a fixed false-positive rate of 10%. It was also possible to assign patient-specific posterior risks based on prior odds and a woman’s adjusted AMH level. Accurate measurement and interpretation of AMH can assist in understanding a woman’s reproductive functional health throughout her natural reproductive life, whether or not she is trying to conceive. This is essential to a woman’s overall quality of life. Early diagnosis of reproductive function disorders can help physicians to intervene and possibly prevent or ameliorate comorbidities of PCOS (25). Adjusted AMH MoM levels below 2.2 indicate reduced patient-specific risk of PCOS, whereas higher levels indicate increased risks for PCOS. The actual posterior risks depend on other well-known factors that influence the prior risk.

Importance of Including BMI in the Interpretation of AMH Levels for PCOS

This study found BMI to be an important covariate of AMH and independent of the woman’s age. The inclusion of BMI significantly improved the interpretation of AMH levels when screening for PCOS. As far as we know, the source of AMH is secreted locally by the ovary and is not externally regulated by, for example, pituitary gonadotropin. The rationale underlying this phenomenon is that the secreted AMH would be diluted in the woman’s circulation based on blood volume and that BMI would be a surrogate for blood volume. This is a well-known effect in maternal serum screening during early pregnancy, in which maternal weight adjustment for clinical markers such as α-fetoprotein and human chorionic gonadotropin are routine (26, 27). There are several potential reasons why some reports may not find a significant BMI effect for AMH. Some may find a negative but not statistically significant relationship, as the study might be underpowered. Others group the AMH results from PCOS and non-PCOS women together when performing the regression, and that can confound the relationship. Many of the analyses do not perform a logarithmic transformation to the AMH results prior to analysis, a necessary step given the multiple orders of magnitude over which those results can range. Alternatively, they may be examining a population with a relatively small range of BMIs (e.g., rural Chinese, native Samoans). Finally, weight rather than BMI may be a better predictor of blood volume. For example, a BMI of 25 kg/m² could occur in a 5-foot-tall, 103-pound woman or in a 6-foot-tall, 184-pound woman. These two women have estimated blood volumes of about 3.1 L and 5.1 L, respectively (28).

Clinical Application: PCOS Diagnosis

AMH correlates strongly with antral follicle count (AFC) and has long been considered a potential marker of polycystic ovarian morphology (PCOM) to replace transvaginal ultrasound (TVUS) for determining AFC. The quality of imaging by TVUS is quite variable due to differences in operator skill, advances in equipment technology, and differing technique. Consensus recommendations advise against its use in diagnosing PCOS during adolescence (postmenarchal age 8 years or less), because of the high incidence of multi-follicular ovaries during this developmental stage. Serum AMH measurement offers promise as an effective, inexpensive, and objective alternative to assess a woman’s antral follicle pool.

However, the new international evidence-based guideline for the assessment and management of polycystic ovary syndrome recommended against using serum AMH measurement as an indicator of PCOM in the diagnosis of PCOS (25). The guideline stated that the primary reason for this conclusion was the lack of an established universal cutoff level for AMH suggestive of PCOM. They relied on a meta-analysis of 29 studies using various assays, life stage, differing PCOS diagnostic criteria, variable ultrasound criteria, and an acknowledged lack of well-characterized normal control subjects. However, the authors do note that AMH could become a more accurate tool in the detection of PCOM with improved standardization of assays and/or by establishing method-specific cut-off levels based on large-scale population-based studies with subjects of different ages and ethnicities (25).

Limitations of the Study

Limitations of this study include the use of banked opportunistic samples taken from a previous study and represent women in the reproductive age range (18–39 years). Samples had been frozen between 2 and 17 years. This study used the NIH criteria to diagnose PCOS-affected women. These criteria include chronic anovulation with clinical and/or biochemical hyperandrogenism, but not PCO morphology on ultrasound. Alternatively, subjects could have been defined in accordance with the Rotterdam criteria (29), which includes TVUS data (i.e., ≥12 antral follicles in one ovary (FNPO) or ≥10 cm³ ovarian volume). It is difficult to speculate how using these alternative criteria would have influenced our analyses, but this could be evaluated in future studies. Also, there are other known covariates of AMH such as smoking and dietary fat intake (30) that were not available in our dataset and other biochemical measurements (Table 1) that are associated with PCOS risk. Further refinement of the AMH MoM for such covariates as gynecological age, race/ethnicity, and diet has the potential to improve this model’s performance as a screening test for PCOS; the addition of other risk factors to a comprehensive model would be a worthwhile future goal, but is beyond the scope of this report.

Strengths of the Study

Although the samples were stored for between 2 and 17 years, there is strong evidence that AMH is stable over such time periods (31). Conversion to MoM prior to modeling has the advantage that it will account for proportional differences between AMH assays. Thus, if one assay is consistently higher (or lower) than another by 20%, there is no need to modify the parameters reported here. If the difference or bias observed between assays are nonproportional, then this is not true, and new modeling is needed. Regardless, revalidation of the algorithm when differences such as a shift in population medians is observed would be reasonable. Additional strengths of the study include the use of widely accepted and reliable modeling techniques to estimate patient-specific risks for PCOS; well-characterized cases and controls; and a reasonably large population to study that allows for confident estimates.

Conclusions

We conclude that converting AMH result to age- and BMI-adjusted MoM levels increases the separation of the two populations, and that subsequent modeling allows for the computation of personalized risk of PCOS. Retrospective as well as prospective validation of this and other AMH applications and risk algorithms for PCOS are warranted. This is especially true for larger-scale community-based studies using varying definitions of PCOS.