Analysis of contraceptive discontinuation in six developing countries from durations of use at survey

Abstract

We conduct a cross‐national study of contraceptive discontinuation among currently married nonsterilized contracepting women in Bolivia, Egypt, Kenya, Sri Lanka, Thailand, and Zimbabwe using the Demographic and Health Surveys (DHS). Since the DHS contains no true completed epochs of contraceptive use, the distribution of use times at survey is used to approximate the distribution of the completed epochs using the renewal theorem. Two techniques based on this approximation are used. The first technique uses local linear regression smoothing of a histogram estimate of the use time at survey pdf which is converted into an estimate of the discontinuation probability function. The second technique poses a proportional hazards Weibull distribution for the discontinuation probability function which is then converted into a model for the use times at survey. This second technique is used to model the observed variations in use across countries while controlling for other sociodemographic factors such as children ever born, age, and education, as well as a variable which encodes knowledge of the fertility cycle. Pill discontinuation probabilities range from 0.12 to 0.47 in the first year. IUD discontinuation probabilities range from 0.18 to 0.53 in the first year. Discontinuation probabilities in Egypt over all methods are in agreement with those reported in Ali and Cleland (1995). Logged relative risks of pill discontinuation range from ‐0.94 (Sri Lanka) to 0 (Kenya), while logged relative risks of IUD discontinuation range from ‐0.53 (Sri Lanka) to 0.41 (Zimbabwe). The ordering of risks of pill discontinuation among the six countries considered is in agreement with the ordering of total fertility rates excerpted from Westoff (1991).