Examining Determinants of IUD Use in India: An Exploratory Analysis of the National Family Health Surveys (NFHS-4) using Machine Learning

Arnab Dey, Pre-Doctoral Fellow
Nabamallika Dehingia, Center on Gender Equity and Health, UC San Diego
Nandita Bhan, Center on Gender Equity and Health, UC San Diego
Edwin Thomas, Center on Gender Equity and Health, UC San Diego
Lotus McDougal, University of California, San Diego
Sarah Averbach, Center on Gender Equity and Health, UC San Diego
Julian McAuley, .
Abhishek Singh, International Institute for Population Sciences (IIPS)
Anita Raj, University of California, San Diego

While India is making progress towards achieving fertility goals, contraceptive use patterns remain lopsided and sterilization-focused, with limited use of contraception to delay or space pregnancies. IUDs are an effective contraceptive method, available free or at low cost in primary care, with low discontinuation rates, but uptake among married women is low (1.5% per recent NFHS). We used machine learning approaches to explore the determinants and barriers to IUD use in India. Data from 499,627 married women were analysed where the outcome of interest was IUD/PPIUD use; all variables in the dataset were included in the machine learning models using logistic regression (lasso and ridge) and neural network approaches along with qualitative coding. Findings revealed that couple’s joint decision-making on family size and contraception as the strongest predictor of IUD use, demonstrating the importance of couple engagement in family planning interventions rather than the singular focus on women. Other key determinants included family planning counselling and contraceptive access, and access to maternal and child services, emphasizing the importance of an integrated approach to women’s health services. Findings also confirm the role of socioeconomic determinants, especially wealth and education in continuing to determine the access and use of family planning services.

Keywords: Family planning and contraception, Big data / Social media

See extended abstract.

  Presented in Session P14.