Clinician vs. self-assessment of mFG hirsutism scores in PCOS: can patient self-assessment serve as a reliable screening and triage tool?

Abstract

Objectives

The aim of this study was to examine the discrepancy between patient- and clinician-assessed modified Ferriman–Gallwey (mFG) hirsutism scores across four PCOS phenotypes and nine body sites, and to evaluate whether patient self-assessment could serve as a reliable tool for preliminary hirsutism screening and outpatient triage.

Methods

This prospective observational study included 157 Han Chinese female patients aged 18–36 years, who were diagnosed with PCOS according to the Rotterdam criteria at an outpatient gynecological endocrinology clinic. All participants completed a self-assessment questionnaire, scoring terminal hair growth across nine body areas using the mFG scale. These self-assessed scores were compared with clinician-assessed scores independently obtained by two trained physicians who were blinded to patient self-assessment results.

Results

Patient self-assessed mFG scores were significantly higher than clinician-assessed scores in the overall PCOS cohort and phenotypes A and B (all P < 0.05), while no significant score differences were detected in phenotypes C and D. Despite higher self-assessed scores, Cohen’s kappa coefficients ranged from 0.551 to 0.855 across phenotypes, indicating moderate to almost perfect diagnostic agreement. Analyses for phenotype D were purely descriptive due to inadequate statistical power. Bland–Altman analysis demonstrated minimal mean scoring bias across all groups, yet the wide 95% limits of agreement (LoA) indicated poor precision for individual-level assessment. The borderline range of self-assessed mFG (5–10), where the cumulative concordance rate fell below the 90% threshold, showed a rate of only 70.6% and was the primary zone of concern. By contrast, patients with scores less than 5 and more than 10 had a concordance rate greater than 90%. ROC curve analysis verified the excellent diagnostic performance of self-assessment (overall AUC = 0.922), with clinician-assessed scores used as an internal reference.

Conclusion

For Han Chinese women with PCOS, self-assessed mFG scoring is a reliable tool for preliminary hirsutism screening and outpatient triage but cannot fully replace clinician assessment. We recommend confirmatory clinician review for patients with self-assessed mFG scores in the borderline zone.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12905-026-04727-1.

Introduction

Polycystic ovary syndrome (PCOS) is the most prevalent gynecological endocrine disorder among women of reproductive age [1, 2], and hirsutism is one of its core clinical manifestations. The modified Ferriman–Gallwey (mFG) score is a widely accepted method for assessing hirsutism, although the hirsutism threshold varies across ethnic populations [3]. The mFG scoring system is routinely completed by clinicians during outpatient consultations [4]. However, in-clinic clinician-assessed mFG scoring is time-consuming in busy practice settings and may compromise outpatient workflow efficiency [5–7]. Therefore, validation of patient self-assessed mFG scoring may help address this clinical limitation by facilitating streamlined outpatient screening and triage.

Studies have reported conflicting conclusions regarding hirsutism self-assessment in African American and White women [8, 9]. Some authors discourage self-assessment due to the risk of misclassification or overestimation of hirsutism severity [8, 10], whereas others recognize its practical clinical value [11–13]. Such inconsistent findings across ethnic groups may be partially attributed to inherent differences in baseline hair follicle density, ethnic-specific hair growth characteristics and variations in body perception and cosmetic awareness among different populations [14–17]. These racial disparities limit the generalizability of Western data to East Asian cohorts [18]. However, studies directly comparing clinician- and patient-assessed mFG scores remain scarce in Han Chinese women, as verified by our PRISMA-compliant literature search (Supplemental Fig. 1). Furthermore, most existing studies merely focused on score differences, whereas few have explored whether such discrepancies affect the diagnostic accuracy of hirsutism.

This study aimed to evaluate discrepancies in hirsutism scores between clinician- and patient-assessed scores across four PCOS phenotypes, and to investigate whether these differences impact hirsutism diagnosis. Additionally, we sought to validate the reliability of patient self-assessed mFG scoring and to explore its potential value in optimizing outpatient screening and triage efficiency. Our findings may provide evidence-based guidance for more convenient and efficient hirsutism evaluation among Han Chinese women with PCOS.

Methods

This prospective observational study was conducted at the Gynecological Endocrinology Clinic of Peking University People’s Hospital from September 1, 2021, to January 1, 2022. The study complied with the 1964 Declaration of Helsinki and its later amendments. Ethical approval was obtained from the Scientific and Ethics Committees of Peking University People’s Hospital (Approval No. 2020PHD007-001). Written informed consent to participate was obtained from all individual participants included in the study. The study enrolled 157 Han Chinese women of reproductive age (18–36 years) diagnosed with PCOS. Demographic and clinical data were collected, and mFG scoring was performed.

Sample size calculation

Sample size calculation was performed based on the primary outcome of paired mFG scores between clinician assessment and patient self-assessment. Given the non-normally distributed nature of the paired data, the estimation was conducted within the statistical framework of the Wilcoxon signed-rank test. According to previous literature, the mean difference was 4.66 (SD = 6.29) [8] which was used as the estimate of the standard deviation of paired differences. A difference of 2.0 points was considered clinically meaningful based on previous literature and clinical relevance [7]. The significance level α was set at 0.05 (two-tailed), and the statistical power (1 − β) was set at 0.80. An asymptotic relative efficiency (ARE) of 0.864 was adopted for ordinal nonparametric sample size correction. The minimum required sample size was estimated to be 93 participants. To ensure adequate statistical precision for agreement analyses and exploratory subgroup comparisons across PCOS phenotypes, and to account for potential attrition, a total of 157 participants were ultimately enrolled. Subgroup analyses by PCOS phenotype were considered exploratory and not independently powered.

Diagnostic criteria

1. Inclusion criteria:

   All patients met the Rotterdam criteria for PCOS, defined by the presence of at least 2 of the following 3 characteristics:

   1. Oligo‑anovulation (OA): Menstrual cycle length < 21 or > 35 days, or < 8 cycles per year, from 3 years after menarche to perimenopause [2].
   2. Biochemical or clinical hyperandrogenism (HA): Biochemical HA: total testosterone (T) ≥ 2.6 nmol/L; Clinical HA: mFG score > 4 [7, 19].
   3. Polycystic ovary morphology (PCO) on pelvic ultrasound ≥ 12 follicles measuring 2–9 mm in diameter in one or both ovaries, and/or ovarian volume ≥ 10 mL (ovarian volume = 0.5 × longitudinal diameter × transverse diameter × anteroposterior diameter) [20].
2. Exclusion criteria:

   1. Adolescent females aged < 18 years;
   2. Hypothalamic amenorrhea, premature ovarian failure, or other conditions causing menstrual irregularities;
   3. Congenital adrenal hyperplasia, androgen-secreting tumors, Cushing’s syndrome, or other hyperandrogenic disorders;
   4. Use of depilatory creams or any form of hair removal before enrollment;
   5. Use of oral contraceptives or spironolactone within the previous 6 months [1].
   6. Other known causes of hyperandrogenism not otherwise specified.

3. PCOS phenotype Classification Criteria [1, 21, 22]:

   PCOS phenotype A: OA + HA + PCO. PCOS phenotype B: OA + HA. PCOS phenotype C: HA + PCO. PCOS phenotype D: OA + PCO.

Assessment of PCO

PCO was assessed using a GE Voluson E8 ultrasound system with a transvaginal probe at a frequency of 5–9 MHz. The diagnosis of PCO was based on the Rotterdam criteria. All evaluations were independently performed by two senior sonographers, and any disagreements were resolved by consensus following discussion.

Sex hormone level detection

Venous blood samples were obtained from all participants after overnight fasting on days 2–3 of the spontaneous menstrual cycle. For patients with irregular menstrual cycles or amenorrhea, blood sampling was conducted only when all of the following criteria were satisfied: a negative urine pregnancy test; transvaginal ultrasound showing no follicles ≥ 10 mm in both ovaries; and endometrial thickness < 5 mm [23]. Serum levels of total testosterone (T), luteinizing hormone (LH), follicle-stimulating hormone (FSH), estradiol (E₂), prolactin (PRL), and progesterone (P) were measured at the Department of Laboratory Medicine, Peking University People’s Hospital. Total testosterone was quantified using liquid chromatography-tandem mass spectrometry (LC-MS/MS). Hyperandrogenemia was defined as a serum total testosterone level ≥ 2.6 nmol/L, in accordance with criteria established by the Department of Gynecological Endocrinology, Peking University People’s Hospital, based on clinical evidence and local population data [1].

Hirsutism assessment protocol

The mFG score, a previously published and validated scale, was used to assess terminal hair growth in this study [7, 22, 24]. The scale evaluates nine body regions: upper lip, chin, chest, upper abdomen, lower abdomen, upper arms, upper back, lower back, and thighs. Each region was scored from 0 to 4, where 0 indicated no terminal hair growth and 4 represented marked virilizing hair growth. The total mFG score was calculated as the sum of scores across the nine regions, and a total score > 4 was defined as hirsutism [1, 2]. Clinical assessments were independently performed by two attending physicians specializing in gynecological endocrinology who had completed standardized mFG scoring training. Clinicians were blinded to patient self-assessment results, and any discrepancies were resolved by a senior physician. The two clinicians received standardized training before formal evaluation and achieved excellent inter-rater reliability, with a weighted kappa coefficient of 0.93. Prior to independent self-assessment, participants received standardized in-person instruction from trained clinicians, including verbal explanation of scoring criteria and review of standardized pictorial aids for each body site. Participants then independently completed the mFG self-assessment questionnaire without further clinician involvement in the scoring process. The whole guidance procedure was strictly standardized to reduce subjective bias and ensure research reproducibility. Both clinician-assessed and patient self-assessed mFG scores were recorded.

Statistical analysis

Data analysis was performed using IBM SPSS Statistics version 27.0 (IBM Corp., Armonk, NY, USA). R software (v4.4.1) with the pROC package (v1.19.0.1) was used for sample size calculation and post-hoc power analysis. All data were anonymized before analysis. The Shapiro–Wilk test was used to assess the normality of data distribution. The Wilcoxon signed-rank test was applied to compare differences in mFG scores between clinician assessments and patient self-assessments. Spearman’s rank correlation analysis was used to evaluate the correlation between clinician-assessed and patient self-assessed mFG scores. McNemar’s test was used to assess systematic differences in binary hirsutism diagnoses between clinicians and patients, while Cohen’s kappa was used to quantify the strength of paired diagnostic agreement. For subgroups with zero events, McNemar’s exact test was used to assess paired diagnostic differences, and Gwet’s Agreement Coefficient 1 (AC1) was applied instead of Cohen’s kappa to evaluate diagnostic agreement, given the known limitations of kappa statistics under conditions of prevalence imbalance or sparse contingency tables. Bland–Altman analysis was performed to assess agreement between clinician-assessed and patient self-assessed total mFG scores. Mean bias and 95% limits of agreement were calculated to evaluate systematic differences and the extent of agreement between the two assessment methods. Scatter plots with marginal histograms and cumulative diagnostic concordance curves were generated using R software (v4.4.1) and the ggplot2, dplyr, ggExtra packages. Receiver operating characteristic (ROC) curve analysis was conducted to evaluate the discriminative performance of patient self-assessed mFG scores for identifying clinician-defined hirsutism, serving as an internal validation of agreement. The area under the curve (AUC) reflects how well self-assessment aligns with the clinician assessment (reference standard). The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated accordingly. To control the risk of Type I error inflation caused by multiple site-specific and phenotypic subgroup paired comparisons, the Benjamini-Hochberg false discovery rate (FDR) correction was applied. All P-values reported in the paired comparison analysis are FDR-adjusted values. A cutoff value of mFG score > 4 was used for hirsutism diagnosis; this cutoff is validated for Han Chinese populations and may not be applicable to other ethnicities. A P-value < 0.05 was considered statistically significant.

Results

Result 1: baseline characteristics and mFG score distribution of study participants

A total of 157 PCOS patients were enrolled and stratified into four clinical phenotype subgroups (A, B, C, D). The overall mean age of the study cohort was 26.29 ± 4.71 years, and the mean body mass index (BMI) was 23.53 ± 5.56 kg/m². Mean hormonal levels were as follows: T (2.63 ± 1.02 nmol/L); LH (11.76 ± 7.81 IU/L); FSH (6.84 ± 2.84 IU/L); E₂ (61.21 ± 69.30 pg/mL); PRL (17.21 ± 26.57 ng/mL); and P (2.28 ± 6.69 ng/mL). The mean clinician-assessed and patient self-assessed mFG scores across the overall cohort and PCOS phenotype subgroups were as follows: overall cohort (4.23 ± 3.43 vs. 5.18 ± 4.80); PCOS phenotype A (4.75 ± 3.61 vs. 5.98 ± 5.49); PCOS phenotype B (4.70 ± 3.54 vs. 5.73 ± 4.56); PCOS phenotype C (3.67 ± 2.19 vs. 4.50 ± 3.18); and PCOS phenotype D, the non-hyperandrogenic phenotype (1.32 ± 0.89 vs. 1.26 ± 1.56) (Table 1).

Table 1.

Basic information of PCOS patients

Type	Number	Age(SD)	Total testosterone(SD)	LH(SD)	FSH(SD)	E2(SD)	PRL(SD)	P(SD)	BMI	clinician-assessed mFG score	self-assessed mFG score
	(%）	(year)	(nmol/l)	(IU/L)	(IU/L)	(pg/ml)	(ng/ml)	(ng/ml)	(mean(SD))	(mean(SD))	(mean(SD))
All	157(100.00%)	26.29(4.71)	2.63(1.02)	11.76(7.81)	6.84(2.84)	61.21(69.30)	17.21(26.57)	2.28(6.69)	23.53(5.56)	4.23(3.43)	5.18(4.80)
A	55(35.03%)	25.58(4.126)	3.00(1.23)	13.92(6.76)	7.18(2.08)	48.83(37.40)	19.37(40.43)	1.54(4.73)	23.29（5.33）	4.75（3.61）	5.98（5.49）
B	71(45.22%)	26.76(5.46)	2.68(0.81)	10.71(8.45)	6.87(3.50)	74.11(90.92)	16.97(10.40)	2.13(6.71)	24.28(6.11)	4.70(3.54)	5.73(4.56)
C	12(7.64%)	25.67(2.31)	2.64(0.46)	9.53(7.75)	6.36(2.34)	51.69(36.53)	24.09(35.17)	1.38(2.12)	20.61(2.92)	3.67(2.19)	4.50(3.18)
D	19(12.10%)	27.00(4.33)	1.49(0.35)	10.96(7.38)	6.08(2.20)	55.79(59.35)	14.13(7.80)	5.27(11.15)	23.31(4.90)	1.32(0.89)	1.26(1.56)

SD standard deviation

Result 2: paired differences in clinician- and patient-assessed mFG scores

mFG score differences varied across phenotype subgroups. To investigate the statistical significance of these between-rater differences, paired Wilcoxon signed‑rank tests were performed for total and site‑specific mFG scores, with stratification by PCOS phenotype. All reported P-values were adjusted for multiple comparisons using the Benjamini-Hochberg FDR method. Key findings are presented below:

For total mFG scores, patient self-assessed scores were significantly higher than clinician-assessed scores in the overall cohort, PCOS phenotype A, and PCOS phenotype B (overall cohort: mean difference = -0.96 ± 2.48, z = -4.328, P < 0.001; PCOS phenotype A: mean difference = -1.24 ± 2.73, z = -3.022, P = 0.003; PCOS phenotype B: mean difference = -1.03 ± 2.53, z = -3.095, P = 0.002). No significant differences in total mFG scores were detected in PCOS phenotype C and PCOS phenotype D. Detailed site-specific paired comparison results for mFG scores are provided in Table 2.

Table 2.

Paired differences in clinician- and patient-assessed mFG scores

Body parts	All-types			A-type			B-type			C-type			D-type
	Score difference (SD)	z	p	Score difference (SD)	z	p	Score difference (SD)	z	p	Score difference (SD)	z	p	Score difference (SD)	z	p
upper lip	-0.24(0.93)	-3.117*	0.014	-1.80（0.88）	-1.48	0.36	-0.27（0.94）	-2.31	0.08	-0.67(1.07)	-1.93	0.17	-0.05(0.91)	0	1
jaw	-0.13(0.44)	-3.586**	0.013	-0.16（0.46）	-2.46	0.06	-0.18（0.49）	-2.919*	0.02	0.08(0.29)	-1	0.53	0.00(0.00)	0	1
chest	-0.01(0.58)	-0.07	1	0.04（0.42）	-0.63	0.77	0.06（0.75）	-0.61	0.77	0.00(0.43)	0	1	0.0526(0.23)	-1	0.53
upper back	0.09 (0.69)	-1.52	0.361	-0.02（0.71）	-0.21	0.99	0.10（0.74）	-1.03	0.53	0.42(0.51)	-2.24	0.08	0.16(0.50)	-1.34	0.39
Lower back	0.05(0.58)	-1.02	0.528	0.03（0.51）	-0.54	0.78	0.06（0.71）	-0.46	0.77	0.08(0.29)	-1	0.53	0.05(0.23)	-1	0.53
Upper abdomen	-0.09(0.63)	-1.75	0.976	-0.07(0.69)	-0.78	0.7	-0.04（0.66）	-0.54	0.78	-0.42()0.51)	-2.24	0.08	-0.11(0.32)	-1.41	0.36
lower abdomen	-0.34(1.00)	-4.133**	0.013	-0.44(1.12)	-2.737**	0.03	-0.41（0.98）	-3.306*	0.01	-0.08(0.10)	0	1	0.00(0.56)	0	1
the upper arm	-0.20(0.78)	-3.167*	0.014	-0.33 (0.82)	-2.814**	0.03	-0.15（0.84）	-1.5	0.36	-0.25(0.75)	-1.13	0.53	0.00 (0.33)	0	1
thigh	-0.04(0.80)	-0.66	0.768	-0.11(0.99)	-0.75	0.71	-0.04（0.76）	-0.51	0.78	0.00(0.60)	0	1	0.11(0.32)	-1.41	0.36
Total score	-0.96(2.48)	-4.328**	0.013	-1.24(2.73)	-3.022**	0.02	-1.03（2.53）	-3.095*	0.01	-0.83(1.80)	-1.41	0.36	0.05(1.68)	-0.27	0.98

SD standard deviation, Z Wilcoxon signed-rank test statistic, p two-tailed P value, adjusted for multiple comparisons using the Benjamini-Hochberg FDR method

*: Significant at the 0.05 level (two-tailed); **: Significant at the 0.01 level (two-tailed)

Result 3: diagnostic agreement for hirsutism classification between clinician and patient self-assessments

To analyze the paired diagnostic agreement between clinician assessments and patient self‑assessments, we used an mFG score > 4 as the cutoff for hirsutism and performed the McNemar’s test and Cohen’s kappa analysis. McNemar’s tests (including McNemar’s exact test for subgroup D) showed no significant systematic difference in paired diagnostic classification in the overall cohort and phenotypes A, B, C, and D (all P > 0.05). Cohen’s kappa coefficients indicated moderate to almost perfect agreement: 0.691 (P < 0.001) for the overall cohort, 0.855 (P < 0.001) for PCOS phenotype A, 0.551 (P < 0.001) for PCOS phenotype B, and 0.667 (P < 0.001) for PCOS phenotype C. For PCOS phenotype D, all analyses were reported for descriptive purposes only. Given the lack of clinician-confirmed hirsutism cases, this subgroup had insufficient statistical power. Thus, the relatively high agreement reflected by Gwet’s AC1 of 0.883 needed to be interpreted with considerable caution.

Furthermore, despite significant differences in mFG scores between clinicians and patients in the overall cohort, PCOS phenotype A and B, no systematic diagnostic bias was detected, and moderate to almost perfect diagnostic agreement was maintained across the overall cohort and phenotypes A, B, and C (Table 3).

Table 3.

Diagnostic agreement for hirsutism classification between clinician and patient self-assessments

Type	Case(n)	Clinician ratings(n(%))		Self-ratings(n)		McNemar	Kappa
		≤4	＞4	≤4	＞4
ALL	157	92(58.60%)	65(41.40%)	84(53.50%)	73(46.50%)	0.152	0.691
A	55	29(52.73%)	26(47.27%)	27(49.09%)	28(50.91%)	0.625	0.855
B	71	38(53.52%)	33(46.48%)	34(47.89%)	37(52.11%)	0.454	0.551
C	12	6(50.00%)	6(50.00%)	6(50.00%)	6(50.00%)	1	0.667
D	19	19(100.00%)	0(0.00%)	17(89.47%)	2(10.53%)	0.5a	0.883b

Values are n (%). McNemar’s test and Kappa coefficient were used for statistical analysis. 

^a McNemar's Exact Test

^b Gwet's Agreement Coefficient 1

Result 4: limits of agreement analysis and diagnostic performance of self-assessed mFG

To assess the limits of agreement (LoA) and diagnostic performance, Bland–Altman analysis, receiver operating characteristic (ROC) curve analysis, and Spearman’s rank correlation analysis were performed.

Spearman’s rho values indicated strong positive correlations between the two assessments in the overall cohort and PCOS phenotype A, B, and C (overall cohort: ρ = 0.844, P < 0.001; PCOS phenotype A: ρ = 0.882, P < 0.001; PCOS phenotype B: ρ = 0.804, P < 0.001; PCOS phenotype C: ρ = 0.822, P = 0.001). No significant correlation was observed in PCOS phenotype D (ρ = 0.218, P = 0.369); however, given the floor effect from minimal mFG score variation in this non-hyperandrogenic subgroup, this finding should be considered exploratory. For Bland–Altman analysis (clinician-assessed minus self-assessed), the mean differences (mean bias) with 95% LoA were − 0.96 (95% LoA: −3.91 to 5.82) for the overall cohort, − 1.24 (95% LoA: −4.10 to 6.57) for PCOS phenotype A, − 1.03 (95% LoA: −3.93 to 5.99) for PCOS phenotype B, and − 0.83 (95% LoA: −2.70 to 4.36) for PCOS phenotype C, and − 0.05 (95% LoA: −3.35 to 3.24) for PCOS phenotype D (reported descriptively given floor effect in this subgroup); the mean bias was small across all groups, with a maximum absolute value of 1.24. Notably, these 95% LoA ranges were wide compared with the full 0–36 mFG scoring scale, suggesting limited precision of individual-level agreement between the two assessment methods (Table 4).

Table 4.

Agreement and correlation between clinician- and patient- assessed mFG scores

Types	Bland–Altman			Spearman
	Bias	Lower LoA	Upper LoA	ρ	P	95% CI
ALL	0.96	-3.91	5.82	0.844	<0.001	0.790 – 0.885
A	1.24	-4.1	6.57	0.882	<0.001	0.802 – 0.931
B	1.03	-3.93	5.99	0.804	<0.001	0.699 – 0.875
C	0.83	-2.7	4.36	0.822	0.001	0.454 – 0.950
D	-0.05	-3.35	3.24	0.218	0.369	−0.275 – 0.621

LoA limits of agreement, CI Confidence Interval

To explore the clinical implications of the wide Bland–Altman LoA, we examined diagnostic discrepancies between the two assessment methods. Among 157 participants, 133 (84.7%) had consistent diagnostic results, and discordant cases predominantly fell within the borderline score range (Supplementary Fig. 2a). Supplementary Fig. 2b presents cumulative concordance rates stratified by self-assessed mFG scores. We defined good agreement as a concordance rate above 90%. The borderline zone (self-assessed mFG: 5–10) showed a concordance of only 70.6% (n = 51), which was markedly lower than that in patients with scores < 5 (90.5%, n = 84) and > 10 (95.5%, n = 22). This stratified concordance pattern highlights the need for confirmatory clinician review in patients with borderline scores.

Using clinician-assessed mFG as the reference standard, self-assessed mFG showed excellent diagnostic performance for hirsutism based on ROC curve analysis: AUC = 0.922 (sensitivity = 0.877, specificity = 0.826, PPV = 78.08%, NPV = 98.48%) in the overall cohort (Fig. 1a); AUC = 0.960 (sensitivity = 0.962, specificity = 0.897, PPV = 89.29%, NPV = 96.30%) in PCOS phenotype A (Fig. 1b); AUC = 0.860 (sensitivity = 0.818, specificity = 0.737, PPV = 72.97%, NPV = 82.35%) in PCOS phenotype B (Fig. 1c); and AUC = 0.917 (sensitivity = 0.833, specificity = 0.833, PPV = 83.33%, NPV = 83.33%) in PCOS phenotype C (Fig. 1d). It should be noted that the reference standard for this ROC analysis was clinician-assessed mFG scores rather than biological gold standards; the reported AUC values therefore reflect internal agreement between the two assessment methods rather than criterion validity against an objective biological endpoint. Post-hoc statistical power analysis revealed that except for subgroup D (N = 19), which had insufficient power due to its lack of clinician-confirmed hirsutism cases, the statistical power of all remaining subgroups was greater than 0.99.

Fig. 1.

Receiver operating characteristic (ROC) curves evaluating the diagnostic performance of patient self‑assessed mFG scores for hirsutism in Han Chinese women with PCOS, stratified by clinical phenotype. a ROC curve for the overall PCOS cohort; b ROC curve for PCOS phenotype A; c ROC curve for PCOS phenotype B; d ROC curve for PCOS phenotype C. The area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) are presented for each subgroup, with clinician‑assessed mFG scores used as the reference standard for hirsutism diagnosis

Discussion

The prevalence of PCOS in women of reproductive age ranges from 11% to 13% [1, 25]. Hirsutism is a core clinical manifestation of hyperandrogenism in patients with PCOS, and the mFG score is the internationally recognized gold standard for hirsutism assessment in clinical practice [1]. However, the reliability of clinician-assessed mFG scoring is compromised by inter-observer variability [1, 26]. Patients’ self-assessed mFG scoring, if validated as a viable alternative to clinician-assessed scoring, may offer substantial advantages in terms of convenience and time efficiency in outpatient screening and triage settings. Nevertheless, existing studies on the diagnostic agreement between self-assessed and clinician-assessed mFG scores remain inconsistent across ethnic groups, and high-quality evidence specifically focusing on Han Chinese women with PCOS remains scarce. This study aimed to evaluate the consistency and diagnostic performance of patient self-assessed mFG scores versus clinician assessment in Han Chinese women with PCOS (stratified by phenotype), and to determine whether patient self-assessment could serve as a reliable adjunct tool for hirsutism screening and triage.

Our core findings confirmed that despite statistically significant differences in mFG scores between clinician and patient assessments in some PCOS phenotypes, there was no systematic diagnostic bias between the two assessment methods. Patient self-assessed mFG scores showed moderate to almost perfect diagnostic agreement, with notable variability across phenotypes. They demonstrated excellent diagnostic performance for hirsutism, except for patients with borderline mFG scores. These findings support self-assessment as a useful adjunct tool for initial hirsutism screening and outpatient triage. However, given the wide individual-level LoA, self-assessed mFG should not be considered a full substitute for clinician assessment. Patients with borderline scores require confirmatory clinician review.

In this study, we found that clinician-assessed total mFG scores were significantly lower than patient self-assessed scores in the overall cohort and PCOS phenotypes A and B, which was consistent with the findings of previous studies [8]. The potential reasons for this trend may be that patients pay greater attention to their own body hair growth and are more sensitive to subtle changes in hair density and distribution, especially in visible sites such as the chin, upper lip, and lower abdomen [27, 28]. In contrast, no significant score difference was observed in PCOS phenotype D, the non-hyperandrogenic phenotype, which was mainly attributed to the extremely low degree of hirsutism in this subgroup, with mFG scores concentrated in the low range and minimal fluctuation between assessments. Notably, we further analyzed site-specific mFG score differences, and found that score differences were site-specific: self-assessed scores were higher in visible sites with high patient attention, while clinician-assessed scores were higher in concealed sites such as the upper abdomen and upper back, which have rarely been reported in previous studies and have provided additional evidence for site-specific differences in mFG assessment [9].

More importantly, despite the statistically significant score differences in some subgroups, our study confirmed that these differences did not affect the diagnostic reliability of hirsutism. The McNemar’s test showed no clinically meaningful diagnostic bias between clinician- and patient-assessed classifications of hirsutism in the overall cohort and phenotypes A, B, and C. Cohen’s kappa coefficients indicated moderate to almost perfect diagnostic agreement, and Spearman’s rank correlation analysis further revealed strong positive correlations between the two assessment methods. This finding suggests that the observed score differences do not lead to systematic misclassification. The Bland–Altman analysis detailed the score differences between the two assessments. Although the mean bias remained minimal across all subgroups (maximum absolute value of 1.24), the 95% LoA were relatively wide compared with the 0–36 mFG scale and indicated that the precision of individual-level agreement was limited. Further diagnostic discordance analysis revealed that 84.7% of cases were concordantly classified at the mFG > 4 threshold. Stratified concordance analysis quantified that diagnostic agreement drops substantially within the self-assessed mFG borderline range of 5–10. Only approximately 70% of patients in this zone achieved consistent hirsutism diagnosis between self-assessed and clinician-assessed scores, whereas concordance exceeded 90% for patients with lower or higher self-assessed mFG scores. This gradient in agreement reinforces our clinical recommendation that confirmatory clinician review is required for borderline-score patients to mitigate misdiagnosis risk. ROC curve analysis further confirmed the diagnostic performance of patient self-assessed mFG score for hirsutism, with an AUC of 0.922 in the overall cohort. It should be noted, however, that the reference standard was clinician-assessed mFG scores rather than biological gold standards such as androgen biomarker levels or treatment response; the AUC therefore reflects internal rater agreement rather than criterion validity against an objective endpoint. In particular, the negative predictive value (NPV) of self-assessed mFG was close to 99% in the overall cohort, indicating that self-assessed mFG can efficiently exclude patients without hirsutism.

The diagnostic agreement of self‑assessed mFG scoring observed in our study was higher than that reported in most previous studies conducted in non-Chinese populations [8, 9]. Notably, those prior Western-based studies enrolled predominantly multi-ethnic Western cohorts, adopted inconsistent mFG rating criteria, and lacked stratified analysis based on PCOS clinical phenotypes. This finding may provide additional insight into population-specific differences in self-assessment patterns, and may be explained by both inherent biological characteristics and sociocultural perceptual differences across ethnicities. Biologically, East Asian women have been reported to exhibit lower baseline body hair density and milder androgen-related hair phenotypes in previous studies compared with Caucasian populations. This may result in a clearer distinction between physiological hair growth and clinical hirsutism [29–31], enabling patients to identify subtle phenotypic changes more accurately and reduce subjective rating bias. Culturally, East Asian populations tend to hold more refined aesthetic perceptions of visible body hair and pay greater attention to hair distribution in facial and abdominal regions [32, 33]. Such body image perception differences may contribute to greater attention to minor hair growth changes, further improving the consistency between self-rated and clinician-rated mFG scores. By contrast, higher inherent hair density and relatively lower attention to subtle bodily hair changes in non-Chinese populations may contribute to greater self-assessment discrepancy [3, 34, 35]. Collectively, ethnic biological traits and cultural perception shape the reliability of mFG self-assessment. Furthermore, the mFG > 4 cutoff applied in this study is validated specifically for Han Chinese populations and may not be applicable to other ethnic groups. Accordingly, both the diagnostic agreement and the corresponding performance metrics reported here should not be directly extrapolated to non-Chinese populations, and population-specific thresholds should be adopted in future studies involving other ethnicities.

The present study has several notable strengths. It specifically focused on Han Chinese women with PCOS, filling the literature gap arising from the poor generalizability of Western findings to East Asian populations. We included patients across all four PCOS phenotypes for subgroup analysis, providing refined evidence beyond simple overall score comparison. Unlike most previous studies that only compared raw mFG scores, this study further explored whether scoring discrepancies affect the diagnostic accuracy of hirsutism, conferring greater clinical implications. Our design also addressed practical clinical challenges such as time-consuming outpatient evaluation, while standardized training for clinician raters with high inter-rater reliability ensured a stable and credible reference standard. Furthermore, our findings suggest that patient self-assessed mFG scoring may serve as a promising, convenient candidate for preliminary hirsutism screening.

Building on our findings, we propose an outpatient triage model. In the immediate term, patients would complete the mFG self‑assessment at the triage station under the guidance of a trained nurse using standardized instructions. As a longer-term extension, we propose a hypothetical future digital model in which the same pictorial aid and standardized video instructions are delivered via an automated smartphone app or pre‑appointment web portal, serving as a potential surrogate for the in‑person instruction. Based on the self‑assessed mFG score, three parallel care pathways are triggered: ①Scores < 5: Hirsutism is unlikely. The treating physician can perform a brief confirmatory visual review during the consultation, which adds minimal time and safely rules out false negatives. ②Scores > 10: These patients are highly likely to meet the hirsutism threshold. They could be prioritized for earlier consultation, and androgen‑related laboratory tests could be pre‑ordered (in accordance with institutional protocols). ③Scores 5–10 (borderline): This zone carries the greatest diagnostic uncertainty. A full, formal clinician‑administered nine‑site mFG evaluation is required for these patients to ensure diagnostic accuracy. This stratified workflow is expected to meaningfully improve triage efficiency in gynecological endocrinology outpatient settings. It is important to note, however, that the digital extension constitutes an exploratory direction. Whether digitally delivered instruction can achieve equivalent diagnostic agreement requires dedicated prospective validation before clinical adoption.

Limitations

This study has several limitations that should be acknowledged. First, the single-center design and modest sample size (n = 157) limit the generalizability of our findings. Furthermore, subgroup analyses of PCOS phenotype D were statistically underpowered and exploratory in nature; all relevant data are reported descriptively, and definitive conclusions cannot be derived exclusively from this phenotype. Large-scale multicenter prospective trials are therefore needed to externally verify our outcomes. Second, enrolling 157 patients within four months may raise concerns about temporal representativeness. However, PCOS-related hirsutism showed no seasonal variation, and no practice changes occurred during the study period. The paired design comparing clinician and patient assessments is inherently robust to temporal bias. Thus, the short recruitment window is unlikely to have substantially affected the main conclusions of this study. Third, we solely enrolled patients with confirmed PCOS and did not include healthy controls or women with other hyperandrogenic disorders, which prevented us from evaluating the diagnostic performance of self-assessed mFG scores in the general population. This aspect could be explored in future expanded studies. Additionally, free testosterone and the Free Androgen Index (FAI) were not assessed in this study, as androgen parameters served solely as phenotype classification criteria rather than analytical variables. Future studies with comprehensive androgen profiling, including FAI and sex hormone-binding globulin (SHBG), may provide additional endocrine context for interpreting phenotype-specific differences in mFG self-assessment patterns. Fourth, clinician-assessed mFG scores served as the reference standard for analyses. Although clinician assessment is the internationally accepted gold standard for hirsutism scoring, mFG scoring is inherently subject to inter-observer variability [36]. To minimize this bias, all clinicians completed standardized training, with an inter-rater kappa value of 0.93 indicating excellent agreement. Even so, subtle residual variability may still exist. Furthermore, the use of clinician-assessed mFG scores as the reference standard for ROC analysis introduces circular reasoning, as both the index test and the reference standard are subjective visual scoring methods. Neither has been validated against biological endpoints such as androgen biomarker levels or treatment response outcomes. Future studies incorporating objective biological criteria as the reference standard are needed to establish the criterion validity of self-assessed mFG scoring. Fifth, the Bland–Altman 95% LoA were relatively wide across subgroups, suggesting that self-assessed mFG scores may not fully replace clinician evaluation for patients with scores near the hirsutism diagnostic threshold. Discordance analysis confirmed that diagnostic misclassification was concentrated in the borderline score range (especially among patients with self-assessed mFG scores of 5–10), supporting the recommendation that borderline self-assessed scores should prompt confirmatory clinician review. Sixth, the mFG cutoff of > 4, the diagnostic agreement between patient and clinician ratings, and the corresponding diagnostic performance are only validated for Han Chinese women with PCOS in the present study. These findings may not be directly generalizable to other ethnic populations, and the applicability of this cutoff and diagnostic performance remains unconfirmed for non‑Chinese racial groups. Investigators working with other ethnic populations should apply population-specific thresholds and exercise caution when interpreting findings based on the mFG > 4 criterion used here. Seventh, all self-assessments were preceded by standardized in-person instruction delivered by trained clinicians in a clinical setting. Whether equivalent diagnostic agreement can be maintained with digitally delivered instruction remains unknown. Therefore, the diagnostic reliability demonstrated in this study cannot be assumed to extend to telemedicine settings, and dedicated prospective studies are needed to evaluate the performance of self-assessed mFG scoring under remote or home-based conditions.

Conclusion

This study evaluated self-assessed mFG scoring in Han Chinese patients with PCOS. Patient self-assessment achieved heterogeneous diagnostic agreement across phenotypes and score strata, with reduced concordance in the borderline score range and high concordance at lower and higher self-assessed mFG scores. As an easy-to-implement tool, self-assessed mFG scoring, preceded by standardized instruction, may provide a convenient adjunct approach for hirsutism screening and outpatient triage. Confirmatory clinician review is recommended for patients with borderline self-assessed scores. The present findings provide refined ethnic-specific evidence for mFG self-evaluation, offering practical references for clinical hirsutism assessment in Han Chinese PCOS populations.

Abbreviations

AUC

Area under the curve

BMI

Body mass index

E₂

Estradiol

FSH

Follicle-stimulating hormone

FAI

Free Androgen Index

HA

Hyperandrogenism

LH

Luteinizing hormone

LoA

Limits of agreement

mFG

Modified Ferriman–Gallwey

NPV

Negative predictive value

OA

Oligo-anovulation

P

Progesterone

PCOS

Polycystic ovary syndrome

PCO

Polycystic ovary morphology

PPV

Positive predictive value

PRL

Prolactin

ROC

Receiver operating characteristic

SHBG

Sex hormone-binding globulin

T total

Testosterone

Author’ contributions

Dehui Su and Yujuan Gao (co-first authors) performed data analysis and drafted the manuscript. Xiaowan Sun, Huina Su, and Ruiqiong Ma contributed to study design and conduct. Yanhua Liu, Xiaolin Jiang, Mengyu Zhang, and Tonghao Lei collected data. Cheng Tan and Xin Yang (corresponding authors) supervised the study and revised the manuscript.

Funding

This work was funded by the National Key R&D Program of China (Grants 2023YFC2706000 and 2023YFC2706002) and the Youth Science Foundation Project of the National Natural Science Foundation of China (Grant No. 82401901).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the 1964 Declaration of Helsinki. Ethical approval was obtained from the Scientific and Ethics Committees of Peking University People’s Hospital (Approval No. 2020PHD007-001). Written informed consent was obtained from all participants prior to enrollment.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.