Fig. 4From: Dirichlet process mixture models to impute missing predictor data in counterfactual prediction models: an application to predict optimal type 2 diabetes therapyAnalysis of variable influence on response prediction for a patient at 6-months. The initial response prediction is conditional on age, number of previous and current therapies. Further predictions are conditional on quantile values for HbA1c, BMI, eGFR and ALT. Out of the 4 missing covariates, the collection of HbA1c will, in theory, provide the most information about treatment response for the patient. Patient: Number of Past Drugs [2], Number of Current Drugs [0], Age [75]Back to article page