Abstract
OBJECTIVE To construct and validate a questionnaire for use in diagnosis of polycystic ovary syndrome (PCOS).
DESIGN All participants completed a questionnaire, which asked clinical questions designed to assist in the diagnosis of PCOS, before their appointments with an endocrinologist. Following completion of the questionnaire, the endocrinologist (blinded to the answers) made or excluded a diagnosis of PCOS using clinical criteria and biochemical data as indicated. Questions were then evaluated for their power to predict PCOS, and a model was constructed using the most reliable items to establish a system to predict a diagnosis of PCOS.
SETTING An outpatient reproductive endocrinology clinic in Calgary, Alta.
PARTICIPANTS Adult women patients who had been referred to the clinic. Fifty patients with PCOS and 50 patients without PCOS were included in the study.
MAIN OUTCOME MEASURES Demographic information, medical history, related diagnoses, menstrual history, and fertility history.
RESULTS A history of infrequent menses, hirsutism, obesity, and acne were strongly predictive of a diagnosis of PCOS, whereas a history of failed pregnancy attempts was not useful. A history of nipple discharge outside of pregnancy strongly predicted no diagnosis of PCOS. We constructed a 4-item questionnaire for use in diagnosis of PCOS; the questionnaire yielded a sensitivity of 85% and a specificity of 85% on multivariate logistic regression and a sensitivity of 77% and a specificity of 94% using the 4-item questionnaire. Predictive accuracy was validated using a second sample of 117 patients, in addition to internal validation using bootstrap analysis.
CONCLUSION We have constructed a simple clinical tool to help diagnose PCOS. This questionnaire can be easily incorporated into family physicians’ busy practices.
Polycystic ovary syndrome (PCOS) is a metabolic disorder characterized by hyperandrogenism and insulin resistance. It is the most common endocrinopathy affecting premenopausal women, with a prevalence of approximately 4.6%.1
Previously there were no widely accepted diagnostic criteria for PCOS. However, a consensus from a conference sponsored by the National Institutes of Health in 1990 determined that the criterion standard diagnosis of PCOS is clinical, defined by the following factors:
-
the presence of ovulatory dysfunction (irregular menstrual cycles and subfertility);
-
the presence of hyperandrogenism (hirsutism or acne); and
-
the exclusion of other related disorders.2
These criteria were recently expanded to include polycystic ovaries apparent on ultrasonography and biochemical hyperandrogenemia, but these criteria are not necessary for diagnosis.3
Polycystic ovary syndrome presents a diagnostic challenge4 to family physicians because of the controversy that has surrounded the diagnostic criteria and because the presenting complaints in PCOS are variable. Most often, patients present with menstrual dysfunction, oligomenorrhea, or infertility5; they can also present with a pregnancy-related complication, such as gestational diabetes6,7 or spontaneous abortion.8,9 Hirsutism or acne could be the patient’s primary concern, which can result in profound psychological distress.8
Polycystic ovary syndrome is associated with several comorbid conditions, including type 2 diabetes,10 dyslipidemia,11 hypertension,12 hepatic steatosis, obstructive sleep apnea,13 endometrial carcinoma, and potentially breast and ovarian cancer.14 It is important to diagnose PCOS as early as possible in the course of disease so that screening, education, and appropriate preventive action and treatment of these patients can be initiated.
To our knowledge, there are no validated tools available in the literature to assist in making the clinical diagnosis of PCOS. We constructed and validated a simple questionnaire for use in screening women for the possible presence of PCOS.
METHODS
Study population
We recruited unselected white patients 18 years or older from an endocrinology reproductive clinic in Calgary, Alta, between January and June 2003. There were no exclusion criteria for participants. The main reasons for referral to this clinic are menstrual irregularity, fertility concerns, and hirsutism. All participants provided written informed consent, and the Conjoint Health Research Ethics Board of the University of Calgary approved the protocol.
Study protocol
Patients were asked to complete the 2-part questionnaire before their appointments with the endocrinologist. The first component requested general demographic information and a medical history, including specific questions regarding known diagnoses of diabetes, hypertension, and dyslipidemia.
The second component of the questionnaire requested a menstrual and fertility history. Patients were instructed to answer these questions excluding time spent pregnant or using pharmaceutical contraception. Questions concerned frequency of menses; history of failed attempts at pregnancy; and history, sites, and treatment of coarse midline hair growth and acne. Patients were asked about a history of breast discharge, a history of obesity, and variability of symptoms with changes in weight.
Once patients completed the questionnaire, the endocrinologist completed the assessment for the criterion standard diagnosis of PCOS (according to the National Institutes of Health criteria). This endocrinologist was blinded to patients’ answers on the questionnaire.
Statistical analysis
Statistical analyses were carried out using Stata, version 8.2. Baseline characteristics of study patients were summarized in terms of frequencies for categorical variables and ranges (mean ± SD) for continuous variables. Bivariate analysis was conducted to assess the association of the predictor variables with the outcome variable of PCOS diagnosis. The Fisher exact test was used for categorical variables, and unpaired t tests were used for continuous variables.
The sample size calculation was powered at 80% to detect a relative risk of 2.5 for a positive response to an item among patients with PCOS relative to patients without PCOS, at an α of .05. This sample size also ensured that the precision of 95% confidence intervals around the sensitivity and specificity of our measure would be no wider than ±10%, provided that our observed values for sensitivity and specificity were 85% or greater.
Logistic regression modeling was used to examine the relationship between patient predictor variables and the outcome of PCOS versus the outcome of no PCOS. All significant (P < .05) baseline predictor variables and interaction terms were used to obtain the backward stepwise selection for the multivariable model. Correlations among the predictors included were checked to avoid colinearity. The final model was assessed by the area under the receiver operating characteristic curve. The goal was to maximize the sensitivity and specificity of the final tool.
Bootstrap analysis was employed to estimate the bias in the predictive accuracy of the model.15 For each bootstrap sample, patients were drawn randomly, with replacement, from the original data set. For each of the 1000 bootstrap samples, the model was then refitted on each bootstrapped data set, with the results inspected for consistency using the bias-corrected confidence intervals for sensitivity and specificity.
Following construction of the model and simplified questionnaire, the questionnaire was issued to a second sample of patients in the same clinic for validation. Sensitivity and specificity were calculated on this validation sample.
RESULTS
Demographic characteristics
A total of 100 subjects participated in the initial phase of the study. Fifty subjects had PCOS and 50 did not have PCOS by the criterion standard. The following diagnoses were established for patients without PCOS: 19 had hypothalamic amenorrhea, 18 had hyperprolactinemia, 5 had premature ovarian failure, 3 had hypopituitarism, 1 had adult-onset congenital adrenal hyperplasia, 1 had idiopathic hirsutism, 1 had menstrual irregularity not yet diagnosed, and 2 were not seen because of menstrual or fertility concerns. Patients with PCOS had a higher average body mass index and a higher prevalence of hypertension than women without PCOS had (Table 1), but the groups did not differ with respect to other demographic characteristics (Table 2).
Patient characteristics
Patients’ education levels and history of illnesses: Apparent discrepancies in percentage values are due to a few patients failing to answer some questions.
Menstrual and fertility history
Significantly more PCOS patients reported a history of long or variable menstrual cycles than patients without PCOS did (36/48 vs 14/49, P = .001) (Table 3). More women with PCOS reported a history of obesity than women without PCOS did (37/48 vs 11/49, P < .001). Patients with PCOS were more likely to report a history of increasing menstrual irregularity with weight gain than those without PCOS were (14/46 vs 3/48, P = .003).
Menstrual and reproductive history
There was no difference in proportion of each group who had previously been pregnant (Table 3). There was no difference between groups with respect to a reported history of 1 year of failed conception attempts. Significantly more patients without PCOS reported a history of nipple discharge outside of pregnancy than women with PCOS did (22/49 vs 3/50, P < .001).
Significantly more women with PCOS reported coarse hair growth than women without PCOS did. Women with PCOS reported hair growth at more of 8 possible sites than women without PCOS did (3.7 ± 2.5 vs 0.8 ± 1.7, P < .001). Women with PCOS who reported hair growth were more likely to report feeling troubled by their hair growth and to have sought treatment for the hair growth than women without PCOS did. More women with PCOS reported that hair growth increased with weight gain than women without PCOS did.
A history of acne was more common among women with PCOS than among women without PCOS (27/50 vs 15/50, P = .03). However, there was no difference in the proportions of each group who had taken prescription treatment for acne.
Predictive model development
Several of the highly predictive variables were considered for inclusion in the model. All factors with a P value <.05 were included in the multivariate analysis. No interaction terms were found to be significant. Four variables (history of obesity, history of long or variable menses, coarse hair growth reported at 3 or more sites, and history of nipple discharge) were included in the final model. The predictive strength of the fit was 0.94 (determined by area under the receiver operating characteristic curve). When a cutoff probability of .45 is used to indicate PCOS, the model has a sensitivity and specificity of 85.4%. Results from the bootstrap analysis showed minimal bias, as indicated by a bias of 2.9% for the sensitivity (bias-corrected 95% CI 63.6%–94.1%), and a bias of 0.8% for the specificity (bias-corrected 95% CI 64.0%–96.3%).
Because the clinical application of a logistic regression model requires calculating probabilities, a cutoff value was selected and significant variables were simplified to develop a scoring system for use in clinical practice. As the coefficients for each item are essentially equal, an equal weighting was assigned to each item (Table 4). The scoring system is a simple sum of each of the 4 items (Table 5). The fourth item regarding a history of nipple discharge generates a negative score, as this item supports a diagnosis other than PCOS. A score of 2 or higher is required for a positive result for PCOS; a score of 1, 0, or −1 represents a negative result. When reapplied to the sample, the sensitivity of the scoring system is 77.1% (95% CI 62.7%–88.0%) and the specificity is 93.8% (95% CI 82.8%–98.7%).
Generation of prediction model and coefficients of variables: Parameter estimates of the logistic regression model.
Clinical tool for diagnosis of polycystic ovary syndrome
Questionnaire validation
The questionnaire was validated by issuing the modified 4-item questionnaire to a second sample of 117 patients at the reproductive endocrinology clinic, 41 of whom had been diagnosed with PCOS by criterion standard. In this sample, sensitivity for the diagnosis of PCOS was 85.4% (95% CI 71.6%–93.1%) and specificity was 93.4% (95% CI 85.5%–97.2%).
DISCUSSION
We have constructed and validated a simple case finding tool that can help physicians diagnose PCOS and can guide them in treating menstrual irregularity, infertility, and cosmetic concerns. This tool can also alert clinicians to screen for associated and potentially devastating comorbid conditions.
This tool has been developed among women whose primary complaint is infertility. Many clinical symptoms among these patients have substantial overlap. For example, women with hyperprolactinemia often present with secondary amenorrhea,16 as do women with PCOS. This selection bias in the referral patient population is likely also reflected in similarity of fertility rates between women with PCOS and women without PCOS. Despite similarities in clinical presentation among women, however, this questionnaire was still able to discriminate between various disease processes with high sensitivity and specificity. Although this tool has not been formally validated in a family medicine clinic, it could discriminate between PCOS and no PCOS even better among women in this population, where primary complaints are often more heterogeneous than in a reproductive endocrinology clinic.
This model includes a history of obesity as a predictor of PCOS, as a history of obesity was strongly predictive of PCOS in our patient population. Although obesity is prevalent among women with PCOS and exacerbates the clinical manifestations of PCOS,13 it must be emphasized that obesity is not essential for the diagnosis of PCOS. Polycystic ovary syndrome is a disorder of excessive androgen production, which is often aggravated by associated insulin resistance.17 Although insulin resistance is closely associated with obesity, it can also manifest clinically in lean patients. The prevalence of obesity among PCOS women ranges from 30% to 75%.13,18 In our population, 52% of women with PCOS were obese.
We included a history of nipple discharge in our clinical prediction tool, as a history of nipple discharge was strongly predictive of a diagnosis other than PCOS. This could reflect selection bias in our population; that is, patients with elevated prolactin levels and amenorrhea are frequently referred to reproductive endocrinology clinics for further assessment. Yet previous research shows that, when pregnancy and PCOS are excluded, one third of patients presenting to family physicians with amenorrhea will have pituitary disease or dysfunction.19 Consequently, it is prudent to include nipple discharge as an important negative predictor of PCOS among women with menstrual irregularity.
Use of this tool does not obviate clinical assessment of these patients. The criterion standard for diagnosing PCOS remains clinical assessment by an expert in the field. This diagnostic tool has been developed using the criterion standard for comparison, however, and thus serves as a reliable case finding tool. A positive result must prompt a careful clinical assessment for metabolic and neoplastic complications of PCOS. A negative result does not rule out PCOS with certainty; in situations of doubt, referral to a reproductive endocrinologist is prudent.
Construction of this questionnaire is subject to some limitations. The sample size of 100 on which the tool was based and the limited number of categories our simplified tool uses to predict outcome restrict our ability to estimate the sensitivity for this measure. We believe that the simplicity of this clinical tool outweighs these limitations, and we hope that future research with this tool will provide a more accurate assessment of its validity.
CONCLUSION
We have constructed and validated a simple clinical tool that is highly sensitive and specific for a diagnosis of PCOS. This questionnaire can be easily used in family physicians’ busy practices.
Notes
EDITOR’S KEY POINTS
-
This validated questionnaire can be useful for screening women with menstrual irregularities, hirsutism or other related findings for the presence of polycystic ovary syndrome. The questionnaire, however, has not been validated in a family medicine setting.
-
A positive score should prompt careful clinical assessment for the metabolic and neoplastic complications of polycystic ovary syndrome.
POINTS DE REPÈRE DU RÉDACTEUR
-
Ce questionnaire validé peut être utile pour dépister la présence du syndrome des ovaires polykystiques chez les femmes ayant des menstruations irrégulières, de l’hirsutisme ou d’autres constatations connexes. L’outil n’a cependant pas été validé dans un milieu de médecine familiale.
-
Un score positif devrait déclencher une évaluation clinique rigoureuse pour détecter les complications métaboliques et néoplasiques du syndrome des ovaires polykystiques.
Footnotes
-
This article has been peer reviewed.
-
Contributors
Drs Pedersen, Faris, and Corenblum contributed to study concept and design, analysis and interpretation of data, and preparing the article for submission. Ms Brar contributed to analysis and interpretation of data and preparing the article for submission.
-
Competing interests
None declared
- Copyright© the College of Family Physicians of Canada
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.