PREDICTIVE MODELLING: PREDICTING HOSPITALISATION AND ESTIMATING THE COST AND RISK TO THE THIRD PARTY FUNDER R Celliers (Msc. (Mathematical Statistics) University of Pretoria) OBJECTIVES: In the third party funder environment most analyses focus on retrospective claims analyses; the aim of predictive modelling is to estimate future claims or current risk, based on the probability of a hospital event using historical data. METHODS: A logistic regression approach is followed where the likelihood of a hospitalisation event is established and mapped to a cost estimate. The modelling process involved establishing a development and validation sample, identifying the predictor variables, building and lastly validating the sample. During the model building process the development and validation population consisted of 149 416 and 47623 beneficiaries respectively ; where the data was obtained from a third party funder consisting of 3 years of data. During the building process the dependant variable (Y=Logg(odds)) takes on the value 1 or 0 depending on whether or not a hospital event occurred.  To calculate the cost per beneficiary a weighted probability was multiplied by the average cost of a hospital authorisation. Beneficiaries were classified as high risk if log(odds)>=0.7. RESULTS: The final model is:  Log(odds)=-0.11X+0.00576X+0.409X+0.179X+0.614X where X-X denotes the predictor variables. X denotes an indicator variable for gender,  X the predictor variable for age, X the HIV indicator variable, X  the diabetes indicator variable and X the chronic indicator variable. The strongest predictors were the chronic indicator variable and age. The validation process resulted in 79% of the beneficiaries being correctly classified and the cost estimation resulted in totals within 3-5% of the actual values. CONCLUSIONS: The proposed model predicts hospitalisation efficiently at a beneficiary level and can be implemented to monitor risk and the associated cost (hospital or total) for individuals, employer group or the third party funder. Third party risk management and cost estimation are other applications of the model.