Markov Modeling Techniques for Selecting Morbidity

Background - Researchers have struggled to develop algorithms to classify important morbidities for inclusion in analyses of public health data. While some morbidities may be quite common, others are relatively rare, and criteria for their inclusion in analyses of morbidity are often not straight forward. To assist with the selection process, an objective methodology by which to determine inclusion criteria for morbidity is warranted. Methods – Using data from the Florida Linked birth certificate and hospital inpatient discharge data (1998-2006), we developed a 3 stage hierarchical model to establish a cut-off point for selecting proportion of morbidities for inclusion in statistical analyses. The Bayesian approach was used for implementation of the model and because of the intractability of the posterior distribution, the Markov Chain Monte Carlo (MCMC) method was used to simulate direct draws from the posterior distribution. The median of the draws of the proportion of morbidities from the posterior distribution was used to establish a cut-off point for morbidity selection. The WinBUGS framework (version 1.4) was used for the analyses. Results – Of the 28 infant morbidities included in our Markov modeling technique, 7 morbidities were selected for epidemiologic analyses based on the mean of the pooled morbidity curve (mean = 0.05). Each of these morbidities had a prevalence of 5% or greater in our study population. Conclusion – The MCMC method which is mostly useful when the posterior distribution is intractable provides a medium for obtaining objective cut-off values for selection of pregnancy-related morbidity of major impact.