Development and Validation of a Machine Learning Algorithm for Problematic Menopause in the Canadian Primary Care Sentinel Surveillance Network (CPCSSN)

Abstract

Abstract Background Menopause is a normal transition in a women’s life. For some women, it is a stage without significant difficulties; for others, menopause symptoms can severely affect their quality of life. Identifying problematic menopause is essential to study the condition and to improve quality of care. This study developed and validated a case definition for problem menopause using Canadian primary care electronic medical records. Methods We used data from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN). A case definition was developed using a reference set created by expert reviewers and a machine learning approach was applied to produce a case definition. Methods to select the most appropriate features and to re-balance our cohort were also applied. Results We randomly selected 2,776 women aged 45–60 for this analysis. An algorithm of two occurrences of ICD-9-CM code 627 in diagnosis fields within 24 months OR one occurrence of ATC code G03CA in medication fields detected problem menopause. This definition produced sensitivity 81.5% (95%CI 76.3%-85.9%), specificity of 93.5% (95%CI 91.9%-94.8%), positive predicted value 73.8% (95%CI 68.3%-78.6%), and negative predicted value 95.7% (95%CI 94.4%-96.8%). Conclusion Our case definition for problem menopause is useful for epidemiological study and demonstrated strong validity metrics. This case definition will help inform future studies exploring management of menopause in primary care settings.