OBJECTIVES:   The objective of this paper is the analysis of practice or physicians'attributes in addition to patient or products 'attributes to explain patterns of drug utilization in predictive disease models including reversed conjoint data. In previous European physicians'cost sensitivity studies, physicians' characteristics such as demographics, solo/group practices, types of remuneration, location (depressed/wealthy areas) were not conclusive (ENDEP/biomed, 2003). This analysis explores additional physicians'attributes more related to professional activities (Rischatsch Zweifel,2012) such as referrals or restrictions to specialists, quality obligations and incident reporting. METHODS: Different types of physicians' choice sets are designed with sets of  attributes  classified in subgroups  Z1,Z2,Z3 ( Z1  for products, Z2 for patients, Z3  for physicians' attributes) . A sample of 688 patients diagnosed with diabetes type II without complications (ICD 250.00) is extracted from the National Medical Care Survey (Huttin/Wong, 2010). Practices are grouped by stages of IT computerization for billing and EMRs. Pharmacological treatment (including oral, injectables and supplies) is defined with a drug list from Facts and Comparisons over three successive years (e.g. 2004/2005/2006) for a 2005 database. Physicians'treatment choices are analyzed with a disease model integrating  Z3 physicians'attributes on subsets of physicians. The cumulative logistic model is run with SAS. RESULTS: Results of the predictive disease model on diabetes type II show on the 2005 analytical dataset that drug utilization is much lower when physicians in clinical practice use a referral process (0.46). The stage of computerization of practices especially with ebilling remains highly significant (lower drug utilization when ebilling is used (0.69)). Additional  runs are tested with predictive models on other chronic conditions (Asthma and Hypertension). CONCLUSIONS: The inclusion of physicians'attributes is critical for discrete choice experiments. This study identifies some statistically significant attributes such as the referral process. It confirms previous results by Miele, Weiland and Dungan (2012) showing how patient can benefit from reduction in HbA1C with centralized referral. Further development investigates how significant physicians' attributes can impact reversed conjoint modeling results on physicians' cost sensitivity.