Handling missing data in modelling quality of clinician-prescribed routine care: Sensitivity analysis of departure from Missing at Random (MAR) assumption