Cost models with prominent outliers

The purpose of this research study is to examine the use of statistical and data mining tools to investigate the funding mechanism for healthcare providers. The specific objective is to examine the relationship between the total charges billed by a hospital compared to the payments received for patient care. Currently, hospitals receive a negotiated payment for a particular diagnosis. The payment is based upon an average cost with local conditions taken into consideration. However, some patients who are severely ill will cost considerably more than average. The question becomes whether hospitals can afford to care for such patients, or whether they are forced to cost-shift to stay in business.;Simulations about the total charges from different distributions were examined to investigate the payment structure. Results from simulations show that when we assume that the distribution of total charges is normal, and the outlier threshold is about two standard deviations from the mean, then the hospitals will almost break even. On the other hand, if the true distribution of total charges is exponential or gamma, then hospitals will lose a considerable amount if the payments are assumed to be normally distributed, and the outlier threshold is about two standard deviations from the mean.;The results were then applied to actual payment data to investigate the cost mechanism. Results show that the distribution of total charges and reimbursements are exponential, or come from an exponential family. Also, there is a significant shift between the two distributions.;We examine the relationship between the total charges billed by a hospital compared to the payments received for patient care using general and generalized linear models. We show the effect of outliers on our linear models. Finally, we compare models with and without outliers. In addition, we show how we can use text mining to reduce a large number of patient condition codes into an index of 4 levels, and to use those levels to examine the relationship to the hospital reimbursements by applying linear models. We compare models with and without the use of a clustering variable to define patient severity.;Predictive modeling tools were used, especially Decision Tree Analysis, in order to investigate the reimbursement model and to try to extract some patterns from the data so these results can be compared to our results using statistical tools to analyze the reimbursement model. The data used in the predictive models are the same data used with linear models, so the comparison between statistical and data mining results can be more accurate and effective. A number of predictors were extracted from the data that might explain the shift of cost and predict the likelihood of future payments.;Results from simulations were confirmed when we used another dataset obtained from a local hospital. The distribution of the total charges for many DRGs chosen was not normal. The hospital loses considerable sums from the outlier payments. The time series were applied on the data to investigate the hospital's total charges and then forecast the future charges in order to help the hospital build a good and effective financial system. Results show that the inpatient's length of stay as a predictor variable has a huge effect, while the hospital's reimbursement doesn't really take this fact into consideration.