Avery Ashby, MS¹, Marisa Allen, MS¹, Soyal Momin, MS, MBA¹, Raymond Phillippi, PhD¹, and Judy Slagle, RN, MPA². (1) Health Services Research, BlueCross BlueShield of Tennessee, 801 Pine St. - 3E, Chattanooga, TN 37402, (423) 763-7482, Avery_Ashby@BCBST.com, (2) Health Info & Reimbursement, BlueCross BlueShield of Tennessee, 801 Pine St. - 9P, Chattanooga, TN 37402

The primary objective of this retrospective cross-sectional study was to develop a predictive model to project future hospitalizations for the diabetic Medicaid population in Tennessee. This model will be utilized to identify and proactively manage diabetic members who are potentially at high-risk for hospitalizations. Information on medical and pharmacy claims was extracted from the clinical data mart called MCSource (Version 5.2) for the period of July 1, 2001 through June 30, 2003. SAS (Version 8.2) was used for data analysis. Data in the first year of the time period was used to model whether a hospitalization occurred during the second year. The study group consisted of 12,683 (mean age 48 years, 64% female) diabetic Medicaid members who were continuously enrolled for the study period, and administered by a MCO in Tennessee. Dual-eligible members were excluded from the study population. Stepwise logistic regression showed that total number of prescriptions (OR = 1.003), number of specialists seen (OR = 1.035), non-insulin users (OR = 0.782), non-antidiabetic drug users (OR = 1.29), number of hospitalizations in previous year (OR = 1.36), number of emergency room visits in previous year (OR = 1.032), non-disabled recipients (OR = 0.813), and metropolitan residents (OR = 0.879) were all related to the risk of a hospitalization in the ensuing year. Chi-square statistics for these variables had p-values of less than 0.01. 79% of the cases were correctly classified and the receiver operating characteristic (ROC) curve showed a fair classification rate (area = 0.703). Comorbidities and demographic variables such as age and gender were not useful in predicting hospitalizations. Utilization measures such as number of hospitalizations in previous year, number of emergency room visits in previous year, and non-drug usage were significant indicators for prospective hospital admissions. Efforts are underway to enhance the predictive model with age related admission weights, lab results, case mix index (CMI) and other quality of care indicators. Data modeling with rational artificial intelligence (RAI) is also planned.

Learning Objectives:

• At the conclusion of the session, the participant (learner) in this session will be able to identify the predictors that sucessfully model hospitalizations for the diabetic Medicaid population in Tennessee.

Keywords: Diabetes, Medicaid Managed Care

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.