Markku Sallmén1, Jens Peter Bonde, Marja-Liisa Lindbohm, Petter Kristensen. 1. From the aFinnish Institute of Occupational Health, Centre of expertise for Health and Work Ability, Helsinki, Finland; b Department of Occupational and Environmental Medicine, Bispebjerg Hospital University of Copenhagen, Copenhagen, Denmark; cDepartment of Occupational Medicine and Epidemiology, National Institute of Occupational Health, Oslo, Norway; and dInstitute of General Practice and Community Medicine, University of Oslo, Oslo, Norway.
Abstract
BACKGROUND: Studies of couple fertility over time have often examined study populations with broad age ranges at a cross-section of time. An increase in fertility has been observed in studies that followed episodes of fertility events either prospectively among nulliparous women or retrospectively among parous women. Fertility has a biological effect on parity. If defined at a cross-section of time, parity will also be affected by year of birth, and thus becomes a collider. Conditioning (stratifying, restricting, or adjusting) on a collider may cause selection bias in the studied association. METHODS: A study with prospective follow-up was taken as the model to assess the validity of fertility studies. We demonstrate the potential for selection bias using causal graphs and nationwide birth statistics and other demographic data. We tested the existence of parity-conditioning bias in data including both parous and nulliparous women. We also used a simulation approach to assess the strength of the bias in populations with prior at-risk cycles. Finally, we evaluated the potential for selection bias due to conditioning on parity in various sampling frames. RESULTS: Analyses indicate that the observed increase in fertility over time can be entirely explained by selection bias due to parity-conditioning. CONCLUSION: Heterogeneity in fertility and differential success in prior at-risk cycles are the ultimate factors behind the selection bias. The potential for selection bias due to parity-conditioning varies by sampling frame. A prospective multidecade study with representative sampling of birth cohorts and follow-up from menarche to menopause would bypass the described bias.
BACKGROUND: Studies of couple fertility over time have often examined study populations with broad age ranges at a cross-section of time. An increase in fertility has been observed in studies that followed episodes of fertility events either prospectively among nulliparous women or retrospectively among parous women. Fertility has a biological effect on parity. If defined at a cross-section of time, parity will also be affected by year of birth, and thus becomes a collider. Conditioning (stratifying, restricting, or adjusting) on a collider may cause selection bias in the studied association. METHODS: A study with prospective follow-up was taken as the model to assess the validity of fertility studies. We demonstrate the potential for selection bias using causal graphs and nationwide birth statistics and other demographic data. We tested the existence of parity-conditioning bias in data including both parous and nulliparous women. We also used a simulation approach to assess the strength of the bias in populations with prior at-risk cycles. Finally, we evaluated the potential for selection bias due to conditioning on parity in various sampling frames. RESULTS: Analyses indicate that the observed increase in fertility over time can be entirely explained by selection bias due to parity-conditioning. CONCLUSION: Heterogeneity in fertility and differential success in prior at-risk cycles are the ultimate factors behind the selection bias. The potential for selection bias due to parity-conditioning varies by sampling frame. A prospective multidecade study with representative sampling of birth cohorts and follow-up from menarche to menopause would bypass the described bias.
Authors: Kirsten J Lum; Rajeshwari Sundaram; Dana B Barr; Thomas A Louis; Germaine M Buck Louis Journal: Epidemiology Date: 2017-01 Impact factor: 4.822
Authors: Emily W Harville; Gita D Mishra; Edwina Yeung; Sunni L Mumford; Enrique F Schisterman; Anne Marie Jukic; Elizabeth E Hatch; Ellen M Mikkelsen; Hong Jiang; Deborah B Ehrenthal; Christina A Porucznik; Joseph B Stanford; Shi-Wu Wen; Alysha Harvey; Danielle Symons Downs; Chittaranjan Yajnik; Donna Santillan; Mark Santillan; Thomas F McElrath; Jessica G Woo; Elaine M Urbina; Jorge E Chavarro; Daniela Sotres-Alvarez; Lydia Bazzano; Jun Zhang; Anne Steiner; Erica P Gunderson; Lauren A Wise Journal: Paediatr Perinat Epidemiol Date: 2019-10-28 Impact factor: 3.980