Sedentary time and stress urinary incontinence in postpartum women: A comparison of isotemporal substitution and compositional data analysis techniques

Vaginal childbirth is one of the most salient risk factors for future pelvic floor disorders, including stress urinary incontinence. As such, it is desirable to find modifiable factors during the postpartum period that may influence a woman's likelihood of developing these pelvic floor disorders. One factor that has been proposed to impact this probability is the amount of sedentary time a woman participates in during the postpartum period. Because a woman's sedentary time is constrained by the number of minutes she is awake, as well as the number of minutes she spends performing light physical activity and moderate-to-vigorous physical activity, statistical techniques that account for these constraints are required. Two such techniques are isotemporal substitution, which involves swapping a set amount of one activity for another activity while keeping the total constant, and compositional data analysis, which transforms the components of a whole from the constrained simplex space to the unconstrained real space. Both techniques were used within a modified Poisson regression model to estimate the prevalence of stress urinary incontinence at one year postpartum based on sedentary time at six months postpartum for first-time mothers. After adjusting for a number of confounders based on a DAG, a statistically significant relationship was not found between sedentary time and stress urinary incontinence using either isotemporal substitution (p=0.6815) or compositional data analysis (p=0.7315). This is likely related to the variables' relationship with BMI at different timepoints, as well as the healing process of the pelvic floor after birth. Although results from isotemporal substitution were easier to interpret, compositional data analysis better accounts for potential asymmetric results when substituting activities for sedentary time and vice versa. An improved method for this type of analysis involves components of both isotemporal substitution and compositional data analysis.