Differences in healthcare expenditure estimates according to statistical approach: A nationwide claims database study on patients with hepatocellular carcinoma

Aim: Disease-associated healthcare expenditures are generally calculated using matched comparisons or regression-based analyses, but little is known about their differences in estimates. This aim of this study was to compare the differences between disease-associated healthcare expenditures estimated using these 2 methods. Methods: In this retrospective cohort study, a matched comparison was first conducted by matching cases with controls using sex, age, and comorbidities to estimate disease-associated expenditures. The cases were then used in a fixed-effects analysis that compared expenditures before and after disease occurrence. The subjects were adults (≥20 years) with primary hepatocellular carcinoma (HCC) who underwent treatment (including surgical resection, locoregional therapy, transcatheter arterial chemoembolization, and transarterial embolization) at a Japanese hospital between April 2010 and March 2018. We calculated the total healthcare expenditures per patient per month according to treatment and disease phase (initial, continuing, and terminal). Results: There were 14,923 cases in the initial/continuing phases and 15,968 cases in the terminal phase. In the initial/continuing phases, 3,552 patients underwent surgical resection only, with HCC-associated expenditures of $5,555 according to the matched comparison and $5,889 according to the fixed-effects analysis (proportional difference: 94.3%). The initial phase expenditures were approximately 9% higher in the fixed-effects analysis, whereas the continuing phase expenditures were approximately 7% higher in the matched comparison. The expenditures in the terminal phase were 93.1% higher in the fixed-effects analysis. Conclusions: The 2 methods produced similar estimates of HCC-associated healthcare expenditures in the initial/continuing phases. However, terminal phase expenditures were substantially different between the methods.