[Efficacy prediction of acupuncture and moxibustion combined with conventional treatment for PCOS-IR based on real-world data: machine learning and interpretability analysis].

Could Acupuncture Help Your PCOS? New Research Uses AI to Predict Who Benefits Most

Polycystic ovary syndrome (PCOS) affects millions of women worldwide, and when paired with insulin resistance (IR), it can make managing weight, blood sugar, and menstrual cycles especially challenging. Many women are turning to acupuncture and moxibustion alongside conventional medical treatment — but could doctors predict in advance who is most likely to benefit? That's exactly what researchers set out to discover.

In this study, scientists collected real-world data from 416 women diagnosed with PCOS and insulin resistance. Using artificial intelligence tools called machine learning models, they analyzed patterns in patients' health data to predict who would respond best to a combined treatment of acupuncture, moxibustion, and standard medical care.

The researchers found that five key health markers were the most important predictors of treatment success: fasting insulin levels (FINS), total cholesterol (TC), the longest length of a menstrual cycle (UML), body mass index (BMI), and a liver enzyme called alanine aminotransferase (ALT). Of these, fasting insulin — a direct measure of how much insulin is circulating in the body — was the single strongest predictor of whether acupuncture and moxibustion would work well.

The AI model, known as a random forest model, performed with strong accuracy and reliability, meaning these five markers together could meaningfully forecast a patient's likely response before treatment even begins.

What does this mean for you? If you have PCOS with insulin resistance, your doctor may one day use a simple panel of blood tests and health measurements to estimate how well acupuncture and moxibustion might work for you — helping personalize your care from the very start.

If you're interested in exploring acupuncture for PCOS, speak with a licensed acupuncturist or integrative medicine provider experienced in women's health and hormonal conditions.

This real-world study developed and validated a machine learning–based efficacy prediction model for PCOS with insulin resistance (PCOS-IR) treated with acupuncture-moxibustion combined with conventional therapy. Training data comprised 284 cases (Jan 2023–Sep 2024); validation set included 132 cases (Sep 2024–Feb 2025). Five algorithms were evaluated; the random forest (RF) model demonstrated the most balanced performance across accuracy, precision, F1 score, and AUC metrics. Feature selection via logistic regression and RF identified five predictive variables: fasting insulin (FINS), total cholesterol (TC), upper limit of menstrual cycle length (UML), BMI, and ALT. SHAP interpretability analysis confirmed FINS as the dominant contributor to outcome prediction, consistent with its central role in IR pathophysiology. Model performance was assessed via ROC curve, calibration curve, and decision curve analysis. Clinically, this model offers a practical framework for prospectively identifying PCOS-IR patients most likely to benefit from integrative acupuncture-moxibustion protocols, supporting individualized treatment planning.