ORF02-05 - Development and External Validation of a Frugal, Clinically-Driven PCOS Risk Score for Primary Care Settings.

Background:
Polycystic ovary syndrome (PCOS) is a common condition but remains underdiagnosed, especially in low- and middle-income countries (LMICs). Scores for diagnosis of PCOS are few, and most of them are very time-consuming and need lots of hormonal and biochemical parameters. Existing AI-based diagnostic models can help in early diagnosis of PCOS, but they are costly and depend on imaging, hormonal assays, and other indicators, which are impractical for use in resource-limited regions. We developed a Frugal PCOS Risk Score using only low-cost, clinical parameters, which can be used in a primary care setting.

Methods:
A total of 100 specialist-audited PCOS and non-PCOS cases were used to construct the model. Univariable and multivariable logistic regression identified younger age, hirsutism, and moderate-to-severe acne as high-yield predictors. These were converted into a simple integer-based score (Frugal Score II). Internal validation was performed using ROC analysis. External validation utilized the full Kaggle PCOS dataset, harmonized for matching clinical variables. Performance metrics included AUC, sensitivity, specificity, and optimal cut-points.

Results:
The Frugal Score II demonstrated strong discrimination in the derivation cohort (AUC = 0.784, 95% CI 0.694–0.873; p < 0.001). Sensitivity-specificity trade-off showed clinically relevant cutoffs, with a threshold ≥1.5 yielding a sensitivity of 62% and a specificity of 83%, suitable for screening. External validation on the Kaggle dataset showed consistent performance, confirming generalizability and stability across populations. The model requires no laboratory parameters and no imaging and can be implemented in less than 3 minutes in primary care settings.

Conclusion:
The Frugal PCOS Risk Score is a low-cost, interpretable, less time-consuming screening tool capable of early PCOS risk identification in primary care settings. The score is ready for deployment as a mobile or web-based app to support primary care clinicians.

Keywords: PCOS, frugal AI, early diagnosis, LMICs, screening tool.