Mining Longitudinal Epidemiological Data To Understand A Reversible Disorder

Medical diagnostics are based on epidemiological findings about reliable predictive factors. In this work, we investigate how sequences of historical recordings of routinely measured assessments can contribute to better class separation. We show that predictive quality improves when considering old recordings, and that factors that contribute inadequately to class separation become more predictive when we exploit historical recordings of them. We report on our results for factors associated with a multifactorial disorder, hepatic steatosis, but our findings apply to further multifactorial outcomes. (1 2)