My methodologic research is to develop statistical models that support scientific learning about human health. My earlier work was on regression models for correlated responses that arise when observations come in clusters, for example in longitudinal research, sample surveys, or when data are observed over time or space. We have extended generalized linear models (logistic, linear, log-linear and survival models) to be applicable in these cases. More recently, my work has been on machine learning algorithms and Bayesian models for "individualized health", that is to use population data to improve decisions about an individual's health state, trajectory or likely benefits and costs of competing interventions. We have applied our novel methods to estimate the etiology of children's pneumonia, trajectory of mental disorders and to predict whether a man's prostate cancer is indolent or aggressive.