Bayesian Hierarchical Approach to Modelling Risk of Miscarriage during First Trimester of Subsequent Intrauterine Pregnancy in Women

Abstract

Background: Accuracy of pregnancy outcome predictions are essential for clinicians to be effective in handling pregnant women experiencing symptoms of miscarriage when presented in the Early Pregnancy Units (EPU). Therefore, the focus of this study is to improve accuracy in modeling risk of miscarriage during first trimester of subsequent intrauterine pregnancy. To achieve this, the study formulates and analyze a Bayesian two-level random intercept logistic model M2 that takes into consideration hierarchical structure in the subsequent pregnancy outcome. This research work is motivated by the recent epidemiological research works that indicated a two-level structure in the subsequent pregnancy outcome with respect to previous pregnancy outcome.

Materials and Methods: The proposed Bayesian hierarchical logistic random intercept model M2 was formulated using the concept of multilevel and Bayesian modeling. Analysis of the proposed Bayesian hierarchical logistic random intercept modelM2 was implemented using the Markov Chain Monte Carlo simulation, in specific Gibbs sampling. Assessment and comparison of the proposed Bayesian hierarchical logistic random intercept model to other reviewed classical approaches using the DIC and AIC.

Results: Assessment of model fitness using DIC and AIC indicates that the proposed Bayesian two-level random intercept logistic modelM2has the lowest value of DIC= 121.2 among both classical and Bayesian approaches M0, M1 and M2. This indicates that the proposed Bayesian model M2 is the most plausible approach in modeling risk of miscarriage during first trimester of subsequent intrauterine pregnancy. Assessment of the between cluster variation as a result of the two-level hierarchical structure in subsequent pregnancy outcome, and prior information which, according to this study, are revealed to be essential in modeling risk of miscarriage during first trimester of the subsequent intrauterine pregnancy in women. Relative biases for the model parameter estimates were all below maximum threshold of |0.2| thus indicating accurate model estimations.

Conclusion: The proposed Bayesian two-level random intercept logistic model M2 is the most plausible approach in modeling risk of miscarriage during first trimester of subsequent pregnancy. Taking into account between previous pregnancy outcome cluster variations and updating observations with prior information is essential to improve accuracy in modeling the risk of miscarriage during first trimester of subsequent pregnancy.