Etsuji Okamoto1, Eiichi Hata. 1. National Institute of Public Health, Department of Management Sciences, 2-3-6, Minami, Wako-shi, Saitama 351-0197, Japan. atoz@niph.go.jp
Abstract
OBJECTIVE: To compare the accuracy and validity of three different methods (Proportional Disease Magnitude method [PDM] with two different magnitude estimations: arithmetic means with correction by the authors; Proportional Allotment Estimator [PAE] by Tango; Maximum Likelihood Estimator [MLE] also by Tango) for estimating disease-specific costs in health insurance claims. METHODS: Application of the three methods to a computer-generated simulation dataset whose disease-specific costs were known and to actual outpatient claims whose disease-specific costs were unknown. OUTCOME MEASURES: For simulation data, the accuracy was assessed by correlation between known disease-specific costs and estimated disease-specific costs by the three methods. For actual claims, concurrent validity was assessed by inter-method correlations between pairs of the two methods. RESULTS: All three methods showed good agreement and accuracy with the simulation data but marked disagreement when they applied to actual claims. MLE yielded an aggregate total of disease-specific costs exceeding the actual total by 21.3% and showed negative disease-specific costs in 18 out of 154 categories. Inter-method correlations showed that PDM with PAE and MLE correlated most strongly (R2 = 0.9022) while the least correlation was observed for PDM with arithmetic means and MLE (R2 = 0.6861). CONCLUSION: MLE is not usable for claims analysis but PDM yielded good estimates with two different methods of magnitude estimation using actual claims.
OBJECTIVE: To compare the accuracy and validity of three different methods (Proportional Disease Magnitude method [PDM] with two different magnitude estimations: arithmetic means with correction by the authors; Proportional Allotment Estimator [PAE] by Tango; Maximum Likelihood Estimator [MLE] also by Tango) for estimating disease-specific costs in health insurance claims. METHODS: Application of the three methods to a computer-generated simulation dataset whose disease-specific costs were known and to actual outpatient claims whose disease-specific costs were unknown. OUTCOME MEASURES: For simulation data, the accuracy was assessed by correlation between known disease-specific costs and estimated disease-specific costs by the three methods. For actual claims, concurrent validity was assessed by inter-method correlations between pairs of the two methods. RESULTS: All three methods showed good agreement and accuracy with the simulation data but marked disagreement when they applied to actual claims. MLE yielded an aggregate total of disease-specific costs exceeding the actual total by 21.3% and showed negative disease-specific costs in 18 out of 154 categories. Inter-method correlations showed that PDM with PAE and MLE correlated most strongly (R2 = 0.9022) while the least correlation was observed for PDM with arithmetic means and MLE (R2 = 0.6861). CONCLUSION: MLE is not usable for claims analysis but PDM yielded good estimates with two different methods of magnitude estimation using actual claims.