Figure 1.

Handling missing data at the time of prediction with different methods.

Consider a scenario where we want to predict SH risk for a patient with SBP and A1C missing. When we only have one (hypothetical) risk model based on age, SBP and A1C, we could handle the missing SBP and AIC by (1) measuring the missing variables, or (2) imputing the missing variables. (3) When we employ the reduced model technique, a reduced model (only containing age as a predictor) of the original model with be used. When we have multiple predictively equivalent risk models available (e.g. model 1-4), we could handle the missing data by (4) find the risk model that matches the data availability of the patient.