Validation of Proportional Distribution Method (Pdm) for Estimating Disease-Specific Costs from Health Insurance Claims: An Empirical Approach

Abstract

Purpose: Proportional distribution method (PDM) is a method to objectively estimate disease-specific costs in a large dataset of health insurance claims usually containing multiple diagnoses but no breaddown of disease-specific costs are included. PDM assumes a common quantity called magnitude for each diagnostic category and distribute the aggregate cost of a claim in proportion to the magnitudes of dianoses contained in the claim. Its validity and accuracy are already demonstrated by simulation and the author attempted to validate it in a more empirical manner. Method: Under Japan's health insurance regulations, outpatient clinics of general hospitals submit an insurance claim for each specialty instead of submitting a single claim aggregating multiple specialties. Outpatient claims submitted from each specialty tend to contain less diagnoses, out of which attending doctors specify a single primary diagnosis. By linking multiple claims of a patient using personal identifiers, aggregate costs with multiple diagnoses were developed and disease-specific costs were estimated using PDM. The results were compared with disease-specific costs calculated by conventional classification by primary diagnoses and their agreement was evaluated by drawing a scatterplot. Data: Outpatient health insurance claims submitted to the National Health Insurance program of Tochigi prefecture (approximately two million population, of whom 883,367 were insured) in June 2005. Total number of outpatient claims was 566,649, of which 97,668 were from general hospitals. Of 97,668 outpatient claims from general hospitals, 40,857 claims were from 18,099 unique patients who visited more than one specialties. The 40,857 claims were aggregated to 18,099 unique patient-level claims with multiple diagnoses. Through the aggregation process, the costs were summed up and the number of visits was assumed to be the largest number of claims of a patient because a patient usually visits more than one specialties on a single visit to hospitals. The resulting 18,099 unique patient-level claims were analyzed by PDM to estimate disease-specific costs. The entire 40,857 claims were also classified by their primary diagnoses and were summed up to yield disease-specific costs. The resulting 119 disease-specific costs estimated by two methods were plotted on a scatterplot for evaluation of their agreement. Results: Disease-specific costs estimated for 119 diagnostic categories by two methods (PDM and conventional classification method) showed a good agreement as evidenced by a regression line (Y=0.9533X+20327) and high R square value (0.96). Conclusion: Validity of PDM had been demonstrated theoretically using simulation. However, it has yet to be validated empirically using actual claims. The results of this study further strengthened the validity of PDM using an empirical model by taking advantage of the peculiar rules of Japan's National Health Insurance program requiring general hospitals to submit outpatient claims by specialties, not as a single claim containing multiple diagnoses from small hospitals or clinics.