525: Early-life prediction of early-onset dysmenorrhoea using machine learning before menarche

Topic: Clinical Science
Subtopic: Pain category (ICD-11), Chronic primary visceral / pelvic pain, Artificial Intelligence/Machine Learning, Epidemiology, Pain in special populations (e.g. neonates, children, teens, elderly, neurodegenerative diseases)

Background and Aims: Dysmenorrhoea in adolescence is common and represents a significant risk factor for the development of chronic pain disorders in women later in life. Identifying girls at risk before menarche could lead to more effective preventive strategies and better pain outcomes in later life. We explored the potential of machine learning models to predict the risk of early-onset dysmenorrhoea in girls before menarche, using prospective data collected during pregnancy and childhood. The potential of ensemble models to outperform individual models in predicting early-onset dysmenorrhoea in girls will also be examined.

Methods: We analysed data from 2703 girls selected from the Avon Longitudinal Study of Parents and Children (ALSPAC), a UK population-based birth cohort, focusing on early-onset dysmenorrhoea within two years after menarche. Maternal prenatal exposures and childhood characteristics assessed through age 7 were included. Candidate predictors spanned psychosocial, anthropometric, medical, and environmental domains. Multiple supervised machine learning algorithms were trained and tested on an independent dataset. Ensemble approaches, including voting and stacking, were applied. Model performance was assessed using balanced accuracy, sensitivity, specificity, F1 score, and area under the ROC and precision-recall curve, with prediction uncertainty quantified using conformal prediction.

Results: Individual machine learning models showed limited predictive performance, with sensitivities being below 50% when specificity was set at 70-80%. However, stacking-based ensemble models were found to have better discrimination capabilities, as evidenced by an increase in the area under the ROC curve to 0.87 and the precision-recall curve to 0.88 in the independent test set. Within this framework, 90.2% of adolescents who developed early-onset dysmenorrhoea were correctly identified using a stacking model with 42 early-life variables. Using prediction uncertainty analysis, it was found that 29.2% of individuals were false positives and 9.8% were false negatives.

Conclusions: The machine learning stacking-based approach has shown promise for predicting early-onset dysmenorrhoea before menarche using early-life variables, with potential to improve early risk stratification and prevention. The higher accuracy of stacking compared with individual models suggests that the complex phenotype of pain is difficult to capture using a single method and that different machine learning approaches identify partially independent features in the data. These gains reflect methodological complementarity rather than overfitting, reinforcing the multifactorial nature of dysmenorrhoea risk. However, validation remains limited by the lack of comparable cohorts, highlighting the need for international collaboration to standardise reproductive and pain-related phenotypes.

References: Reid-McCann R, Poli-Neto OB, Stein K, Dixon S, Cox E, Coxon L, et al. Longitudinal association between dysmenorrhoea in adolescence and chronic pain in adulthood: a UK population-based study. Lancet Child Adolescent Health 2025;9(11):766-75.
Dixon S, Hirst J, Taghinejadi N, Duddy C, Vincent K, Ziebland S. What is known about adolescent dysmenorrhoea in (and for) community health settings? Front Reprod Health 2024;6:1394978.
MacGregor B, Allaire C, Bedaiwy MA, Yong PJ, Bougie O. Disease burden of dysmenorrhea: impact on life course potential. Int J Women’s Health 2023;15:499-509.

Presenting Author: Omero B Poli-Neto
Submitting Author: Omero Poli-Neto

Senior Author: Katy Vincent

IASP 2026 Global Year theme of Neuropathic Pain: No

• Rachel Reid-McCann - Nuffield Dept of Women’s and Reproductive Health, University of Oxford, UK.
• Kate Stein - Department of Psychiatry, University of Oxford, UK
• Sharon Dixon - Nuffield Dept of Women’s and Reproductive Health, University of Oxford, UK
• Emma Cox - Endometriosis UK
• Lydia Coxon - Nuffield Dept of Women’s and Reproductive Health, University of Oxford, UK
• Mina Fazel - Department of Psychiatry, University of Oxford, UK
• MaryAnn Noonan - Department of Psychology, University of York, UK
• Gemma C Sharp - Department of Psychology, Faculty of Health and Life Sciences, University of Exeter, UK
• Krina T Zondervan - Nuffield Dept of Women’s and Reproductive Health, University of Oxford, UK
• Katy Vincent - Nuffield Dept of Women’s and Reproductive Health, University of Oxford, UK