Evaluating the PCOS risk algorithm (PriskA): a pilot study.

This study is single-centre study conducted between March 2023 and February 2024. We aimed to include 150 women who attended a standardized endocrine screening at the outpatient clinic of Reproductive Endocrinology and Infertility at the Erasmus University Medical Centre Rotterdam presenting with one or more symptoms of PCOS. After an extensive endocrine screening, the cause of these PCOS-related symptoms was clearly identified. This could have resulted in a diagnosis of PCOS, but also other diagnoses, including WHO 1 (hypogonadotropic hypogonadism), WHO 2 (normogonadotropic anovulation without PCOS), WHO 3 (hypergonadotropic hypogonadism, including early menopause), or healthy controls. After obtaining informed consent from the patient, this diagnosis, which was part of standard healthcare, was compared to the risk score generated by the PriskA. Other inclusion criteria were: age between 16 and 45 years, sufficient command of the Dutch language, and signed written informed consent to participate in this study. Women were excluded if they were pregnant, had a history of malignancy, could not undergo a transvaginal ultrasound, or had used hormonal medication (including hormonal intrauterine devices) in the past three months. This study was approved by the Medical Ethical Review Board of the Erasmus University Medical Centre Rotterdam (MEC-2022-0587). This study was registered at www.clinicaltrials.gov , registration no. NCT05785507 . The endocrine screening was part of standard healthcare to diagnose women with an irregular cycle and/or suspected PCOS or another endocrine disorder. The screening was executed by three different doctors who were well-trained for this procedure. Due to the irregular or absent cycle of these women, this screening is scheduled on a random day, irrespective of the previous menstrual period. The screening was performed in the morning after an overnight fast and included the assessment of the menstrual cycle, height and weight, calculation of body mass index (BMI), measurement of waist and hip circumference, evaluation of hirsutism using the modified Ferriman-Gallwey scale (mFGs), measurement of systolic and diastolic blood pressure, and blood withdrawal. Total follicle count (TFC) and ovarian volume were assessed using transvaginal ultrasound using a transducer > 8 Mhz. The following hormone levels were measured: luteinizing hormone (LH), follicle-stimulating hormone (FSH), estradiol (E 2 ), inhibin B, progesterone, anti-Müllerian hormone (AMH), 17-hydroxyprogesterone, testosterone, androstenedione, dehydroepiandrosterone (DHEA), DHEA-sulfate (DHEAS), cortisol, prolactin, sex hormone-binding globulin (SHBG), thyroid-stimulating hormone (TSH), and fasting levels of insulin and glucose. Androstenedione, DHEA, DHEA-S, testosterone, progesterone, 17-hydroxyprogesterone, and cortisol were measured using a self-developed ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) method. AMH and TSH were measured using Fujirebio Lumipulse G1200 system. SHBG was measured using an immunoassay (IDS-iSYS). FSH, LH, estradiol, prolactin, and insulin were measured using Siemens Atellica IM1300 system. Glucose was measured using the Cobas ® 8000 Modular Analyzer (Roche Diagnostics GmbH). Inhibin B was measured using an ELISA immunoassay (Beckman coulter Gen II). For the validation of the PriskA algorithm, AMH, testosterone and SHBG were determined using Elecsys AMH Plus, Elecsys Total Testosterone and Elecsys SHBG on a Cobas 6000 Modular Analyzer (Roche Diagnostics GmbH, Germany). The criteria for PCOS diagnosis were according to the 2018 and 2023 International Guideline for PCOS, including oligomenorrhea (menstrual cycle  35 days or  182 days), hyperandrogenism (clinical or biochemical hyperandrogenism), and polycystic ovarian morphology (PCOM) [ 6 , 12 ]. Clinical hyperandrogenism was defined as a modified Ferriman-Gallwey score (mFGs) of 5 or greater. Biochemical hyperandrogenism was defined as a total serum testosterone level > 2.0 nmol/L or a free androgen index (FAI = total testosterone/SHBG x 100%) > 2.9. PCOM was defined as 20 or more follicles (2–9 mm in diameter), and/or increased ovarian volume (> 10 mL) in at least one ovary, evaluated by using a transvaginal ultrasound transducer with a frequency > 8 MHz. Since this was a prospective study, women were included in the study before the laboratory results were available. Therefore, biochemical hyperandrogenism, as part of the laboratory results, could not be used as an inclusion criterion in this study. The diagnosis of PCOS was initially based on the 2018 International Guidelines for PCOS, using ultrasound for the diagnosis of PCOM without the use of Anti-Müllerian hormone (AMH). For the second analysis, an AMH cut-off of 3.2 ng/mL (using the Roche Elecsys ® AMH Plus assay) was applied to detect PCOM instead of ultrasound data. This was done because AMH appears to be an adequate representation of the number of follicles in an ovary, and has been included in the new 2023 PCOS guideline [ 12 , 13 ]. This AMH cut-off for PCOM using the Elecsys AMH Plus immunoassay was derived and validated in the previously published APHRODITE study, as no universally accepted cut-off is available yet [ 14 ]. This cut-off was later also validated in another prospective study [ 15 ]. The PriskA was developed by Roche Diagnostics and based on clinical data, including age, BMI, menstrual cycle information (regular, irregular, or absent), and levels of AMH, testosterone, and SHBG (using the Roche Elecsys AMH Plus, Total Testosterone and SHGB immunoassay). Initially, LH levels were assessed, and if LH was found to be lower than 2, a low-risk score was assigned due to a highs suspicion of suppression or hypogonadotropic status. If LH was greater than 2, the PriskA algorithm was applied. This algorithm is a logistic regression equation and included laboratory values measured by the Elecsys method on a Cobas 6000 analyser (Fig.  1 and Supplemental Fig. 1). Therefore, these hormone values were measured twice: once for standard clinical care of the Erasmus University Medical Centre (using immunoassay established at Erasmus University MC) and once for the PriskA tool (using Elecsys immunoassays mentioned above). For the latter, participants were required to provide an additional blood sample to enable the specific measurements needed for the study. Both sets of hormone values were measured simultaneously during the same blood withdrawal, with the additional sample taken solely for the PriskA-specific measurements. Fig. 1 PriskA algorithm PriskA algorithm The PriskA tool generated a risk score ranging from 0 to 1. A score below 0.2 was considered to indicate a low risk of having PCOS, a score between 0.2 and 0.8 indicated a moderate risk, and a score above 0.8 indicated a high risk of having PCOS. The algorithm was developed based on a training set of more than 1600 controls and over 1800 subjects. In this training, PriskA achieved an area under the curve (AUC) of 98%. In this pilot study using real-world data, we expected to achieve an AUC of 0.70 and performed a power calculation, which indicated that a total of 107 patients (including cases and controls) would be required. Considering potential data loss due to exclusions or blood withdrawal failures, our target was to include 150 patients in total (approximately 100 PCOS patients and 50 non-PCOS patients). To generate the PriskA score via the algorithm, a digital tool was developed. Three healthcare providers who conducted the standard endocrine screening during this period used this tool. After entering the parameters, the tool displayed the final PriskA score. At the end of the study, the three healthcare providers received a questionnaire on the usability of this digital tool and the validity of the PriskA. Since this is a pilot study including only three healthcare providers, this questionnaire was an exploratory feasibility assessment. This questionnaire, which focuses on the primary algorithm, asked users to rate various usability aspects on a scale from 0 to 5. Continuous variables were compared between women with and without PCOS using the Mann-Whitney U test. Categorical variables were compared using the chi-square test. A p-value of < 0.05 was considered statistically significant. To evaluate the validity of the PriskA tool, sensitivity and specificity were assessed using the standard-of-care diagnosis (PCOS or no PCOS) and the low and high PriskA scores in ROC analyses. ROC curve analysis was also used to calculate the AUC for the PriskA algorithm (as a continuous score) and individual biomarkers (AMH, T, FAI, LH, SHBG). DeLong’s test was performed to compare the AUCs of the PriskA algorithm with each biomarker to determine whether the combined PriskA score provides significantly better discrimination between PCOS and non-PCOS cases. A p-value of < 0.05 was considered statistically significant. Model performance was evaluated using the F1-score, which is the harmonic mean of precision and recall. Precision was calculated as the number of true positives divided by the sum of true positives and false positives, and recall as the number of true positives divided by the sum of true positives and false negatives. Bootstrap validation was performed to assess the stability and variability of the model’s performance.

Of the 150 women initially included in the study, six were excluded due to an inconclusive diagnosis at the time of data analysis ( n  = 2 hyperprolactinemia, n  = 3 hypothyroidism, n  = 1 suppressed LH and FSH levels) (Fig.  2 ). These patients may also have had PCOS, but this diagnosis could only be made after addressing these endocrinological issues and were therefore excluded. Of the 144 remaining women, 95 were diagnosed with PCOS, while 49 did not have PCOS but exhibited only one PCOS feature, as this was part of the inclusion criteria. Among the women without PCOS, the following diagnoses were made: WHO 1 (hypogonadotropic hypogonadism) ( n  = 5), WHO 2 (normogonadotropic anovulation) without PCOS ( n  = 25), early menopause/perimenopause ( n  = 8), regular cycle with solely PCOM or hyperandrogenism ( n  = 7), endometriosis ( n  = 1), temporary amenorrhea following leuprorelin use ( n  = 1), and suspected congenital adrenal hyperplasia (CAH) ( n  = 1). Age was similar between the PCOS group and non-PCOS group (PCOS 29.7 (IQR 26.7–33.1) versus non-PCOS 31.7 (IQR 27.1–34.4), p  = 0.12) (Table  1 ). Women with PCOS had a higher BMI (26.6 (IQR 22.6–31.9) versus 23.6 (IQR 21.4–27.5), p  < 0.01) and had more often ovulatory dysfunction (97.9% versus 85.7%, p  < 0.01). Levels of AMH and testosterone, and the FAI were significantly higher in women with PCOS (all p-values < 0.01). SHBG was significantly lower in women with PCOS ( p  < 0.05). The mFGs and TFC was significantly higher in the PCOS group (3.0 (IQR 1.0–6.0) versus 1.0 (0.0–2.5) and 46.0 (IQR 33.0–58.0) versus 21.5 (IQR 11.3–31.8), both p-values < 0.01). In the PCOS group 53 (55.8%) had phenotype A, 12 (12.6%) phenotype B, 1 (1.1%) phenotype C, and 29 (30.5%) phenotype D. Fig. 2 Flowchart patient inclusion. Abbreviations : PCOS Polycystic ovary syndrome, LH Luteinizing hormone, FSH Follicle-stimulating hormone, WHO 1 Hypogonadotropic hypogonadism, WHO 2 Normogonadotropic anovulation, CAH Congenital adrenal hyperplasia Flowchart patient inclusion. Abbreviations : PCOS Polycystic ovary syndrome, LH Luteinizing hormone, FSH Follicle-stimulating hormone, WHO 1 Hypogonadotropic hypogonadism, WHO 2 Normogonadotropic anovulation, CAH Congenital adrenal hyperplasia Table 1 Baseline characteristics Standard of care diagnosis PCOS ( n  = 95) No PCOS ( n  = 49) P -value Variables used for PriskA algorithm  Age 29.7 (26.7–33.1) 31.7 (27.1–34.4) 0.12  BMI 26.6 (22.6–31.9) 23.6 (21.4–27.5) < 0.01  Ovulatory dysfunction (%) 93 (97.9) 42 (85.7) < 0.01  AMH (ng/mL, Elecsys) 6.0 (3.8–7.8) 2.2 (1.1–3.7) < 0.01  Testosterone (nmol/L Elecsys) 1.5 (1.0–2.0) 1.2 (0.6–1.4) < 0.01  SHBG (nmol/L, Elecsys) 40.6 (25.3–55.1) 48.6 (34.6–63.7) < 0.05  FAI (Elecsys) 3.8 (2.3–6.0) 1.7 (1.0–2.9) < 0.01 Other variables used for clinical care diagnosis  Testosterone (nmol/L, LC-MS/MS) 1.2 (1.0–1.7) 0.8 (0.6–0.9) < 0.01  SHBG (nmol/L IDS-iSYS) 37.1 (23.0–52.8) 48.2 (34.1–62.0) < 0.05  FAI (LC-MS/MS) 3.6 (2.4–5.2) 1.5 (1.1–2.4) < 0.01  mFGs 3.0 (1.0–6.0) 1.0 (0.0–2.5) < 0.01  TFC 46.0 (33.0–58.0) 21.5 (11.3–31.8) < 0.01 Phenotypes  A (%)  B (%)  C (%)  D (%) 53 (55.8) 12 (12.6) 1 (1.1) 29 (30.5) - - Data are presented as medians with interquartile ranges or as numbers with percentages Abbreviations : BMI Body mass index, AMH Anti-Müllerian hormone, SHBG Sex hormone binding globulin, FAI Free androgen index, mFGs Modified Ferriman-Gallwey score, TFC Total follicle count Baseline characteristics A (%) B (%) C (%) D (%) 53 (55.8) 12 (12.6) 1 (1.1) 29 (30.5) Data are presented as medians with interquartile ranges or as numbers with percentages Abbreviations : BMI Body mass index, AMH Anti-Müllerian hormone, SHBG Sex hormone binding globulin, FAI Free androgen index, mFGs Modified Ferriman-Gallwey score, TFC Total follicle count Of the 49 women without PCOS, 17 had a low PriskA score, 6 a moderate score, and 26 a high PriskA score. Of the 95 women with PCOS, two had a low PriskA score, one a moderate score, and 92 a high score (Table  2 ). Including only high and low scores ( n  = 137), it resulted in a sensitivity of 0.98, a specificity of 0.40 and a F1 score of 0.85 (Table  3 and Supplemental Fig. 2a). The 26 patients with a false positive PriskA score were diagnosed with CAH ( n  = 1), endometriosis ( n  = 1), temporary amenorrhea following leuprorelin use ( n  = 1), perimenopause ( n  = 1), WHO 2 without PCOS ( n  = 20), WHO 1 ( n  = 1). Table 2 PriskA scores versus standard-of-care diagnosis Standard-of-care diagnosis using ultrasound Standard-of-care diagnosis using AMH (cut-off 3.2 ng/mL) No PCOS PCOS Total No PCOS PCOS Total Low PriskA score 17 2 19 17 2 19 Moderate Priska score 6 1 7 6 1 7 High PriskA score 26 92 118 21 97 118 Total 49 95 144 44 100 144 Abbreviations : AMH Anti-Müllerian hormone, PCOS Polycystic ovary syndrome, PriskA PCOS risk algorithm PriskA scores versus standard-of-care diagnosis Abbreviations : AMH Anti-Müllerian hormone, PCOS Polycystic ovary syndrome, PriskA PCOS risk algorithm Table 3 Sensitivity and specificity of PriskA scores Standard-of-care diagnosis using ultrasound Standard-of-care diagnosis using AMH (cut-off 3.2 ng/mL) No PCOS PCOS Total No PCOS PCOS Total Low PriskA score 17 2 19 17 2 19 High PriskA score 26 92 118 21 97 118 Total 43 94 137 38 99 137 Sensitivity 0.98 0.98 Specificity 0.40 0.45 F1 score 0.85 0.87 Sensitivity and specificity of PriskA scores When using the Elecsys AMH Plus cut-off of 3.2 ng/mL for PCOM instead of ultrasound, 7 patients previously diagnosed as not having PCOS were reclassified as ‘PCOS’, while 2 patients previously diagnosed with PCOS were reclassified as ‘non-PCOS’ (Table  2 ). Comparing this new ‘standard-of-care diagnosis’ with the PriskA score yielded a sensitivity of 0.98, a specificity of 0.45 and a F1 score of 0.87 (Tables  3 and Supplemental Fig. 2b). The number of false-positive patients decreased from 26 to 21, including 2 patients who were diagnosed with PCOS based on ultrasound but not with the AMH-based approach. Of the 19 false positive patients left, the diagnoses included suspicion of CAH (n = 2), endometriosis (n = 1), temporary amenorrhea following leuprorelin use (n = 1), perimenopause (n = 1), WHO 2 without PCOS (n = 13), WHO 1 (n = 1). Figure 3 shows the ROC curves of the single markers in the PriskA tool, using standard-of-care diagnosis with ultrasound (A) and with AMH (B). For both standard-of-care diagnosis AMH showed the highest AUC (0.84 and 0.86 respectively). Testosterone showed an AUC of 0.74 and 0.72, FAI 0.74 and 0.72, LH 0.70 and 0.70, and SHBG 0.61 and 0.60. DeLong comparison showed that the PriskA score had the highest AUC (0.89) compared to the variables separately, indicating excellent predictive accuracy. It also showed that the variables T, FAI, LH and SHBG had significant contribution in the algorithm, while the AUC of AMH alone did not differ significantly form the AUC of the PriskA score ( p  = 0.6 for both standard-of-care diagnosis) (Supplemental Tables 1 and 2). Bootstrap validation indicated that, for standard-of-care diagnosis using ultrasound, the estimated sensitivity and specificity had bootstrap variances of 0.0003 and 0.0045, respectively. For diagnosis using AMH, the bootstrap variances of sensitivity and specificity were 0.0002 and 0.004, respectively. Fig. 3 ROC curves of single markers in PriskA tool using standard-of-care diagnosis with ultrasound ( A ) and using AMH B . Abbreviations : AMH Anti-Müllerian hormone, T Testosterone, LH Luteinizing hormone, SHBG Sex hormone binding globulin, FAI Free androgen index ROC curves of single markers in PriskA tool using standard-of-care diagnosis with ultrasound ( A ) and using AMH B . Abbreviations : AMH Anti-Müllerian hormone, T Testosterone, LH Luteinizing hormone, SHBG Sex hormone binding globulin, FAI Free androgen index We observed that the variable oligomenorrhea/amenorrhea was weighted too heavily in the algorithm. Therefore, we developed an improved algorithm which predictive value was explored within the same dataset. This tool is more a decision tree, and included only low- and high-risk scores, without a moderate-risk category (Fig.  4 ). These results should be interpreted as hypothesis-generating, and independent external validation is required before clinical application. Results of the improved PriskA tool compared with the standard-of-care diagnosis (both with ultrasound and with AMH) are shown in Table  4 and Supplemental Figs. 3a and 3b. The improved PriskA showed a sensitivity of 0.97, a specificity of 0.67 and a F1 score 0.89 compared to the standard-of-care diagnosis with ultrasound, with 16 false positives and 3 false negatives observed. When using the standard-of-care diagnosis with the AMH cut-off of 3.2 ng/mL to detect PCOM instead of ultrasound, the improved PriskA demonstrated a sensitivity of 0.98, a specificity of 0.77 and a F1 score of 0.93, with 10 false positives and 2 false negatives. Fig. 4 Improved PriskA algorithm. AMH was measured in ng/mL, LH was measured in U/L. Oligomenorrhea was defined as a menstrual cycle  35 days or  182 days. Abbreviations : LH Luteinizing hormone, OA Oligo- or amenorrhea, AMH Anti-Müllerian hormone, FAI Free androgen index Improved PriskA algorithm. AMH was measured in ng/mL, LH was measured in U/L. Oligomenorrhea was defined as a menstrual cycle  35 days or  182 days. Abbreviations : LH Luteinizing hormone, OA Oligo- or amenorrhea, AMH Anti-Müllerian hormone, FAI Free androgen index Table 4 PriskA scores of the improved algorithm versus standard-of-care diagnosis Standard-of-care diagnosis using ultrasound Standard-of-care diagnosis using AMH (cut-off 3.2 ng/mL) No PCOS PCOS Total No PCOS PCOS Total Low PriskA score 33 3 36 34 2 36 High PriskA score 16 92 108 10 98 108 Total 49 95 144 44 100 144 Sensitivity 0.97 0.98 Specificity 0.67 0.77 F1 score 0.89 0.93 Abbreviations : AMH Anti-Müllerian hormone, PCOS Polycystic ovary syndrome, PriskA PCOS risk algorithm PriskA scores of the improved algorithm versus standard-of-care diagnosis Abbreviations : AMH Anti-Müllerian hormone, PCOS Polycystic ovary syndrome, PriskA PCOS risk algorithm The results of the questionnaire from three researchers about the usability of the PriskA digital tool (using the primary algorithm) are shown in Fig.  5 . The tool received moderate ratings for effectiveness in estimating PCOS risk (average 3.7/5) and for utility in supporting clinical decision-making (average 3.0/5). General practitioners and general gynaecologists were perceived as likely to benefit from the tool (average rating of 3.3), whereas it was deemed less useful for PCOS specialists (average score of 1.7). Additionally, the tool was generally rated as user-friendly. Feedback on the consistency of PriskA scores with the final clinical diagnosis was mixed. The users suggested potential algorithm refinements, such as considering ovulatory function, progesterone, prolactin and thyroid function. Fig. 5 Results of the questionnaire about the usability of the PriskA digital tool. Answers are scaled from 1 to 5 where 1 equals ‘strongly disagree’ and 5 equals to ‘strongly agree’ Results of the questionnaire about the usability of the PriskA digital tool. Answers are scaled from 1 to 5 where 1 equals ‘strongly disagree’ and 5 equals to ‘strongly agree’

The aim of this pilot study was to evaluate the validity of the PriskA digital tool for assessing PCOS risk in a real-world setting. Our findings demonstrate that the original algorithm achieved high sensitivity (0.98) but lower specificity (0.40) when compared with standard-of-care diagnosis using ultrasound to detect PCOM. When substituting AMH for ultrasound in the standard-of-care diagnosis, sensitivity remained high (0.98), with a slight increase in specificity to 0.45. By refining the algorithm based on our collected data, we developed an improved exploratory version of PriskA, which achieved a sensitivity of 0.97 and specificity of 0.67 (using ultrasound in the standard-of-care diagnosis) and a sensitivity of 0.98 and specificity of 0.77 (using AMH instead of ultrasound). These results suggest that PriskA could be a valuable tool for physicians less familiar with diagnosing PCOS such as General Practitioners (GPs) or internal medicine physicians. However, independent external validation is required before clinical implementation. While this tool may offer limited additional value for specialized clinical centres with expertise in PCOS, it holds significant potential for general practitioners or other medical specialists. By enabling earlier referral, it could address the persistent issue of delayed diagnosis and its associated consequences for women. Timely diagnosis not only reduces the burden on patients themselves but also facilitates early screening and management of PCOS-associated comorbidities, such as obesity, type 2 diabetes, dyslipidaemia, hypertension, and metabolic syndrome. To achieve timely diagnosis and early screening, a key feature of a screening tool is high sensitivity, ensuring a low rate of false negatives and minimizing the risk of missed cases. The PriskA tool meets this high sensitivity-criterion with a sensitivity of 98%, making it particularly useful in primary care settings. Although the specificity of 77% may lead to more referrals of women who do not have PCOS, many of these women may still present with other health issues warranting attention. For example, in our study, we identified patients with conditions such as hyperprolactinemia, hypothyroidism, early menopause, and even two patients with suspected CAH. It is important to note that this study was conducted in a tertiary care hospital and therefore involved a referred population. This population is different from the women typically seen in primary care setting, as referral populations often represent more complex or atypical cases [ 9 , 16 ]. Consequently, the positive and negative predictive values of the tool are expected to be lower in primary care populations, where disease prevalence is lower. This difference may also explain the lower validity observed in our study compared to the training set. The training set primarily included phenotype A PCOS cases, which represent the most pronounced form of the syndrome, along with controls that excluded women with irregular menstrual cycles or other endocrine disorders [ 1 , 17 ]. This referral bias highlights a limitation of the current study: the findings may not be fully generalizable to a broader population of women with suspected PCOS. Another limitation is that the ultrasound scans were performed by three different clinicians, and inter-observer reliability, although in general comparable, was not formally assessed. Additionally, the relatively small sample size, despite the required number achieved according to the power calculation, may limit generalizability. On the contrary, a major strength of this study is the prospective evaluation of the PriskA tool in a real-world setting of women with suspected PCOS, enhancing the applicability of the findings to clinical practice. Additionally, the refinement of the algorithm based on the data demonstrates its adaptability and potential for further optimization. By comparing the tool’s performance using both ultrasound and AMH as diagnostic standards, this study provided valuable insights into its utility across different diagnostic approaches. Recent years have seen growing interest in the role of AMH in diagnosing PCOS. While AMH is strongly correlated with follicle count, it has shown limited utility as an independent, stand-alone biomarker for PCOS [ 18 – 20 ]. One likely explanation is the heterogeneity of PCOS, where PCOM is not the only issue and just one of three diagnostic Rotterdam criteria for PCOS and may not always be present, such as in PCOS phenotype B (OA + HA). Combining AMH with other markers that capture the broader spectrum of PCOS could improve diagnostic accuracy and may help in diagnosing women with PCOS more efficiently, and thereby earlier. We have tested the algorithm by comparing it to the established diagnostic criteria as they are currently used. However, it is possible that such an algorithm could ultimately perform better than our current diagnostic methods, potentially identifying cases that are missed or misdiagnosed (for example, the ‘WHO 2 – no PCOS category’ in this study). Another explanation for the limited utility of AMH as a biomarker is the influence of other factors on AMH, including assay platforms, age and BMI [ 21 – 23 ]. It is important to recognize that this algorithm was also developed using AMH based on a specific assay. This algorithm does take age into account, however, more research is needed on the use of AMH in adolescents, as the hypothalamic-ovarian axis is still maturing in this population and PCOM is not uncommon [ 18 ]. Recent studies have explored artificial intelligence and large datasets for predicting PCOS. However, these approaches often require extensive data collection and complex infrastructure, limiting their applicability in clinical practice [ 24 – 26 ]. In contrast, PriskA is designed to function with minimal data inputs, making it a feasible solution in real-world settings. This study, to our knowledge, is the first to validate a digital tool for PCOS risk assessment prospectively, marking an important step toward integrating algorithm-based tools into routine care.

In this pilot study, we demonstrated that the PriskA is a promising tool for assessing PCOS risk in a referral population. The tool demonstrated high sensitivity with minimal missed cases, while specificity significantly improved after exploratory refinements. These results are hypothesis-generating, and predictive values are expected to be lower in primary care populations with lower disease prevalence. Although PriskA shows potential to support earlier identification and proper and timely referral of at-risk women, independent external validation in broader and more diverse populations is required before recommending its use in primary care. By facilitating timely referrals, the tool may help reduce diagnostic delays and improve management of PCOS-associated comorbidities, potentially preventing adverse metabolic outcomes, such as type 2 diabetes. Overall, this study highlights the adaptability and potential clinical applicability of PriskA, representing a step forward in integrating digital tools into routine PCOS care, offering a practical and scalable approach for improving women’s health outcomes.

With a prevalence up to 15%, polycystic ovary syndrome (PCOS) is the most common endocrine disorder in women of reproductive age [ 1 ]. Women with PCOS present with diverse features, including reproductive features such as irregular menstrual cycles, subfertility, hirsutism and pregnancy complications, metabolic features such as obesity, insulin resistance, metabolic syndrome, and type 2 diabetes (all cardiovascular risk factors), and psychological features such as anxiety and depression [ 2 – 5 ]. Because of the reproductive, metabolic and cardiovascular risk factors it is important to screen and inform these women [ 6 ]. However, a considerable number of the affected women remain undiagnosed [ 7 ]. PCOS is a heterogeneous disorder that is defined by a combination of signs and symptoms of androgen access and ovarian dysfunction. According to the Rotterdam criteria in the international guidelines, PCOS is indicated if at least two out of the three of the following criteria are met: ovulatory dysfunction (oligo- or amenorrhea (OA)), clinical and/or biochemical hyperandrogenism (HA) and/or polycystic ovarian morphology (PCOM) [ 8 ]. Following these criteria, PCOS is classified in four different phenotypes depending on the diagnostic criteria that are fulfilled: phenotype A (OA + HA + PCOM), phenotype B (OA + HA), phenotype C (HA + PCOM), and phenotype D (OA +PCOM). The distribution of the phenotypes shows a difference between an unselected population and a referred or hospital-based population. A previously published study reported a distribution in a referral population of 50% phenotype A, 13% phenotype B, 14% phenotype C and 17% phenotype D, whereas in an unselected population the distribution was 19% phenotype A, 25% phenotype B, 34% phenotype C and 19% phenotype D [ 9 ]. This difference in phenotype distribution suggests a selection bias with regards to who ultimately receives a diagnosis, potentially leading to delayed or missed diagnoses of PCOS. This selection bias is compounded by the fact that women with irregular cycles or other PCOS-related symptoms are often the first present to general practitioners (GPs). However, it is particularly challenging for GPs to assess whether a patient has PCOS, as they typically lack sufficient knowledge of hormonal imbalances, particularly in androgens, and are unable to perform ultrasound to assess for PCOM. Furthermore, not only GPs, but also medical specialist lack consensus regarding the latest criteria for PCOS diagnosis. This all results in underdiagnosis, overdiagnosis and misdiagnosis. Additionally, the varying cut-off values used in diagnostic criteria can add to the confusion, as different thresholds can lead to discrepancies in the identification of PCOS, making it even more challenging for healthcare providers to reach accurate conclusions. These diagnostic inaccuracies contribute to feelings of distress and distrust among patients, while also limiting opportunities for effective and timely prevention and intervention for PCOS-associated comorbidities [ 10 , 11 ]. These gaps in early diagnosis underscore the need for improved diagnostic approaches that can more accurately identify patients at risk, particularly in a primary care setting. Therefore, the PCOS risk algorithm (PriskA), a digital tool to assess the risk of PCOS in patients with signs and symptoms of PCOS, has been developed. This tool requires neither ultrasound nor specialized knowledge of hormones, making it usable for all healthcare providers. The aim of this pilot study was to evaluate the validity of the PriskA digital tool in a prospective, real-world setting at a specialist centre.

Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.