The pressing need for standardization in epidemiologic studies of PCOS across the globe

Abstract

The polycystic ovary syndrome (PCOS) is a common and important complex endocrine metabolic disorder affecting women mainly in the reproductive age. The prevalence of the disorder varies depending on the epidemiologic design and criterion used to study the disease. This variation in methodology and subsequent effect on epidemiologic estimate makes it difficult to compare prevalences and phenotypes across geographical areas and assess the effect of cultural and racial variations on PCOS phenotypes. Overall, there is an urgent need for a globally accepted standardized protocol for epidemiologic studies of PCOS, which will maximize the comparability of studies around the globe. To address this issue the Androgen Excess and PCOS Society, Inc. has designated an expert Task Force to draft recommendations to guide epidemiologic research worldwide. Once completed, the use of such recommendations will enable epidemiologists to the effects of geographical and cultural variations of PCOS prevalence and assist in determining the phenotype–genotype associations in the disorder. Further, it will assist in developing informed, and thus effective, public health policy. In essence, the need to standardize epidemiologic studies across the globe is pressing and urgent.

摘要

多囊卵巢综合征（PCOS）是一种常见且重要的复杂内分泌代谢紊乱疾病, 主要影响育龄期女性。该疾病的患病率取决于研究该疾病的流行病学设计和标准。方法学的变化以及对流行病学估计的相继影响使得难以比较不同地理区域的患病率和表型, 也难以评估文化和种族差异对PCOS表型的影响。总而言之, 全球迫切需要公认标准的PCOS流行病学研究方案, 这将最大限度地提高全球研究的可比性。为了解决此问题, 雄激素过量和PCOS 协会已经指定了一个专家工作组来起草指导全球PCOS流行病学研究的建议。一旦完成, 这些建议将使流行病学家能够了解地理和文化差异对PCOS患病率的影响, 并有助于确定疾病中的表型 - 基因型关联。此外, 它将有助于制定知情, 有效的公共卫生政策。从本质上讲, 全球PCOS流行病学研究标准化的必要性迫在眉睫。

Keywords:

• PCOS
• epidemiology
• prevalence
• hirsutism
• androgen excess

Introduction

Polycystic ovarian syndrome (PCOS) is a common endocrine metabolic complex genetic trait affecting women and is most apparent in the reproductive age. The syndrome is characterized by chronic oligo-anovulation (OA), biochemical and/or clinical hyperandrogenism (HA) and polycystic ovarian morphology (PCOM). PCOS impacts negatively reproductive function (infertility, recurrent pregnancy loss, etc.) and predisposes to adverse obstetric outcomes, such as gestational diabetes mellitus, pre-eclampsia, fetal macrosomia, and perinatal morbidity and mortality. There is also a direct link between PCOS and endometria carcinoma, type-II diabetes mellitus (T2DM), and several other metabolic disorders and possibly cardiovascular disease in later years. Dyslipidemia is more common in women with PCOS, with patients demonstrating higher levels of low-density lipoprotein (LDL)-cholesterol and triglycerides, and lower levels of high-density lipoprotein (HDL)-cholesterol compared with women without the disorder.

While PCOS is most obvious clinically in reproductive age women, emerging evidence suggests that prepubertal girls and postmenopausal women may also have related symptoms. Postmenopausal women are at increased risk of T2DM and possibly cardio-metabolic complications though symptoms attributable to excess androgen may improve or even disappear. PCOS in children may manifest as premature pubarche while menstrual irregularity and clinical evidence of hyperandrogenemia may be the main signs and symptoms in the adolescent group.

PCOS prevalence is affected by variations in methodology and research design

Globally, the reported prevalence of PCOS varies from 5% to 20% in various studies, depending on which diagnostic criterion was used (or more precisely which phenotypes were included, see below), how the study population was identified, the methods used to define each feature of the criterion, how complete were the phenotypic assessments, and the recruitment process of the study population [1]. Of note, most if not all studies of PCOS prevalence are from North America, Europe, the Middle East, Southern Asia, and Australia; there is no significant data as of yet from South America, Russia (i.e. Northern Asia), the island countries of Oceania (Melanesia, Micronesia, and Polynesia) and Africa [2]. In fact, our group is working to address the paucity of data arising from the African continent. These variances in methodology create great degree of uncertainty around prevalence and phenotype of PCOS, hampering the comparability of studies; the interpretation of genetic analyses; the ability to detect the impact of race/ethnicity, environment, socioeconomics, and diet and nutrition, among other factors; the estimation of economic burden; and the development of informed and effective public health policy.

Where it has been studied so far, the prevalence of the disease has been fairly constant at 5–9% in most parts of the world when using the stricter 1990 NIH Criterion, which includes two of the PCOS phenotypes (Phenotype A: OA + HA + PCOM and Phenotype B: OA + HA) [1]. When the 2006 Androgen Excess & PCOS (AE-PCOS) society diagnostic criterion is used (which, in addition to Phenotypes A and B, includes Phenotype C, i.e. HA + PCOM) prevalence rates for PCOS of 10–15% are reported. The use of the European Society for Human Reproduction and Embryology (ESHRE)/American Society for Reproductive Medicine (ASRM) definition, more commonly known as the 2003 Rotterdam Criteria (which in addition to PCOS Phenotypes A, B and C, includes Phenotype D, i.e. OA + PCOM), increased the reported prevalence rate for the disorder to as high as 20% in some studies. We should note that phenotypically Rotterdam 2003 and AE-PCOS Society 2006 definitions are effectively expansions of the NIH 1990 criterion. Regardless, the diagnostic criterion chosen affects the prevalence of the disorder observed.

How the study population is identified plays an important role in determining prevalence. For example, the phenotype of the disorder observed in a population is under significant referral bias and there is evidence that PCOS phenotype identified in clinical (referred) cohorts disproportionately represent for more severe forms of the disorder compared with the phenotypes identified from studies of medically unselected populations. For example, Ezeh et al compared two cohorts from the same geographical area, the first consisting of cases seeking medical care and the second consisting of patients identified through a routine pre-employment medical screening [3]. They found that referred cases had a higher prevalence of the more complete PCOS phenotypes (phenotype A), were more hirsute, had higher serum androgen levels and were more obese compared with the medically unselected cohort. This bias, at least in part is drive by the negative impact of the disorder on quality of life, principally around obesity and features of HA and ability to access medical care [3]. A subsequent meta-analysis of all published studies demonstrated the same results [2]. Thus populations used to study the prevalence of PCOS should be medically unselected as possible. These may include the study of randomly selected women from community-based cohorts or, less optimally, women undergoing an assessment for non-medical reasons.

Methodologic issues in defining each feature of the criterion

The methods used to define each feature of the criterion are critical in ensuring the complete detection of the disorder. How OA, HA and PCOM is defined will play a significant role in determining the prevalence of PCOS observed. For example, should OA be defined by the menstrual dysfunction only? Should the degree of oligomenorrhea be defined as cycles (vaginal bleeding episodes) at greater than 35 day intervals, or 45 day intervals (equivalent to 8 or less cycles per year)? Using older epidemiologic data, it seems that cycles >35 days in length are abnormal, which is equivalent to 10 or less cycles per year. However, many studies use <8 cycles per year as the definition of oligomenorrhea, equivalent to cycles >45 days in length. Furthermore, some of the OA may not be detected by overt menstrual dysfunction, and may present as polymenorrhea or eumenorrhea. These oligo-ovulatory patients can be detected only if their late luteal (day 22–24) progesterone levels are assessed.

Even more controversy exists around the definition of HA. Clinically HA is often defined by the presence of hirsutism, but how does the presence of acne and androgenic alopecia fit in to the definition? And how is hirsutism defined? Today the most common method of detecting hirsutism is through the assessment of terminal hair growth in male-like body areas using a visual scale, the modified Ferriman-Gallwey (mFG) score [2]. Less certain is the exact cutoff value to use. In White or Black women, hirsutism has been defined by a value of 6–8 or greater, representing the 95th percentile of the populations studied.

However, a question we should ask is why is a biological variable such as hirsutism being determined by a strict percentile? A measure that indicates that 5% of women studied are ‘abnormal’ and the rest are not. Implying that, by definition, one in every 20 women has hirsutism. The use of this percentile stems perhaps from confusion between what we commonly use as an acceptable degree of statistical error (i.e. 5% or a p < .05) and what should be defined as ‘abnormal’. Or perhaps it stems from an assumption that hirsutism follows a normal Gaussian curve distribution (which it does not) where mean plus two standard deviations encompasses 95% of all measures –but which actually would mean that ‘abnormal’ on the high side would be defined as the upper 2.5%. As if we would define ‘abnormal’ fasting glucose. Blood pressure, cholesterolemia, or body mass using percentiles .

In a large study of unselected black and white women we used cluster analysis and associated symptom to define ‘abnormal’, which our data indicated to be a cutoff value of 3 or more [4]. A study in Han Chinese women found a comparable cutoff value using a similar approach [5]. So should we define hirsutism by an mFG score value greater than 3? Or 6 or 8 (or 10, as some investigators prefer to use)? Obviously, the number of women detected as having clinical HA will vary according to cutoff (the lower the cutoff the more subjects will be discovered).

And even more difficult may be defining HA biochemically. Should we assess total Testosterone (T), free T, DHEA, androstenedione (A4), dihydrotestosterone (DHT)? Others? One, some or all? Evaluating a cohort of PCOS patients diagnosed by the NIH 1990 criterion, total T alone (using a high-quality assay) detected some 33% of patients, the addition of free T measures increased the detection rate to 60%, and the addition of DHEA and A4 increased the detection rate by approximately 7–10% each. And speaking of androgen measures, the quality of assays vary widely. Total T is the mainstay of androgen measures in women, alone and in the estimation of free T and the standard for total T assay is either a high quality radio-immunoassay following sample extraction and column chromatography or, better still, mass spectrometry. Yet the quality of total T varies widely in studies of PCOS and many investigators use total T (generally alone) measured by direct platform immunoassays. In fact, Lizneva and colleagues observed in their meta-analysis that the weakest areas of reporting were related to insensitive androgen measures [2]. So which assays and hormones or prehormones should be assessed in detecting PCOS? And as androgens change with age should age-related cutoff values be used?

Finally, while PCOS is usually detected by ultrasonography, what exact cutoffs should be used? The Rotterdam criterion defined PCOS by an ovarian volume of 10 cm² (ml) and/or an antral follicle (reflected by cysts 2–9 mm in diameter) count (AFC) of 12 or more (Rotterdam, 2004). Finding in one ovary suffices. However, as technology has improved, the ability to detect small follicles has increased. So it is likely that AFC of 12 is too low, and investigators have suggested that AFCs of 18–22 should be used as the lower limit for the diagnosis of PCOS [6]. Alternatively, ovarian volume does not seem to change much with improved transvaginal probe frequency. And what about those patients in whom a transvaginal approach cannot be used? And does age matter? Some investigators have used similar methods (i.e. cluster analysis) to identify women with ‘abnormal’ ovarian morphology as has been used for determining ‘abnormal’ facial or body terminal hair growth or androgen levels in the circulation [6].

And not only do epidemiologic studies of PCOS suffer from the use of different criteria, different recruitment schemes, and even more variability in phenotyping, but the diagnostic scheme for PCOS itself advocates against a complete assessment of these subjects. ‘Healthy controls’ in a population are often diagnosed simply by the absence of a history of irregular menstruation and medical problems, and the absence of clinical signs of HA on physical exam. Alternatively, diagnosing a woman with PCOS requires many more tests. It requires that subjects undergo blood testing to exclude thyroid dysfunction and hyperprolactinemia, and 17-hydroxyprogesterone to exclude 21-hydroxylase deficient nonclassic adrenal hyperplasia. It may even require additional tests to exclude Cushing’s syndrome, congenital adrenal hyperplasia, or androgen-secreting neoplasms. As mentioned, full phenotyping may also require measurement of luteal progesterone levels and/or transvaginal ultrasonography. Hence, there is a much higher likelihood that subjects with PCOS will not be less willing to complete their evaluation, which may take multiple visits and invasive tests, than controls, yielding many more ‘incompletely assessed’ subjects with PCOS than controls (and consequently biasing against the PCOS diagnosis). For example, in a large population-based study of medically unselected women of reproductive age in Tehran, Iran, Tehrani et al reported that, more than a third of their cases would have been undiagnosed or misdiagnosed had they not assessed the participants for subclinical menstrual dysfunction or biochemical hyperandrogenism [7]. We have addressed this evaluation bias by including all subjects in our analysis, whether completely or incompletely evaluated, and assigning those women who were incompletely evaluated a ‘weight’ for the diagnosis of PCOS based on the results in similarly phenotyped individuals who were completely assessed [8].

Conclusions

Overall, these issues strongly indicate that epidemiologic studies of PCOS worldwide vary greatly in methodology and outcome, and comparisons can only be inferred. Overall, there is an urgent need for a globally accepted standardized protocol for epidemiologic studies of PCOS, which will maximize the comparability of studies around the globe. To address this issue the Androgen Excess & PCOS Society, Inc. has designated an expert Task Force to draft recommendations to guide epidemiologic research worldwide. Once completed, the use of such recommendations will enable epidemiologists to the effects of geographical and cultural variations of PCOS prevalence, and assist in determining the phenotype–genotype associations in the disorder. Further, it will assist in developing informed, and thus effective, public health policy. In essence, the need to standardize epidemiologic studies across the globe is pressing and urgent.

Disclosure statement

R. A. consults for Ansh Labs, Longitude Capital, Spruce Biosciences, and Medtronics. He serves on the advisory board of Martin PET Imaging. No potential conflicts of interest was reported for any of the remaining authors.