OBJECTIVES: Retrospective pharmaceutical outcomes studies require controlling observable factors that influence physician choice and patient heterogeneity to minimize selection bias. However, most studies neglect the assessment of physician prescribing habit as a contributor to this choice. This study provides evidence that the physician prescribing habit is an influential factor in determining medication choice. METHODS: A Medicaid claim database was used to study the factors determining the initial prescription choice among 3 classes of asthma controller medications: inhaled corticosteroids, theophylline, and cromolyn. A total of 4748 pediatric asthma patients with an 8-month washout period were selected.A total of 834 different physicians prescribed controllers to this population. Thirty-five covariates were selected to model initial prescription choice, including patient demographics, comorbidities, previous drugs, health costs, seasonality, provider prescribing habit and volume. Physician prescribing habit was defined as the most frequently prescribed controller medication. To ensure exogeneity, physician prescribing habit and volume were defined from a separate population of 24,260 patients with controllers prescribed by the same cohort of physicians. We compared different multinomial logit (MNL) regressions according to the percentage of correct predictions generated from each model. A non-parametric data-partitioning tree (by SPSS/AnswerTree(r)) with Chi-square Automatic Interaction Detector (CHAID) method was applied to confirm the findings. RESULTS: The MNL model containing only one factor, physician prescription habit, correctly predicted 57.4% of the medication choices, while the MNL model with all other covariates only predicted 52.3% correctly. A combination of all 35 covariates achieves a prediction rate of 59.9%. The data-partitioning tree with CHAID method selected prescribing habit as the first variable to classify the outcome tree (chi-square=1367, df=6). Additional covariates identified by the CHAID method included race, prescription volume, and prescription volume squared. CONCLUSIONS: Physician prescribing habit is an influential factor in prescription decision choice in this case, and should not be neglected in retrospective pharmaceutical outcomes studies.