Design and analysis of unit cost estimation studies: How many hospital diagnoses? How many countries?

We evaluated three questions that commonly arise when unit costing exercises for multinational trials are conducted: (1) In countries where investigators plan to collect hospital unit cost estimates for a selected set of diagnoses, how should one estimate unit costs for the remaining diagnoses observed in the trial for which cost data were not collected? (2) For how many hospital diagnoses should estimates be obtained? (3) For how many countries should they be obtained? We addressed these questions using unit cost data collected in four western European countries and three relative value measures from the US Medicare diagnosis-related group (DRG) payment system. We found that the arithmetic mean length of stay from the US DRG payment system was a good predictor of unit costs in four countries in Europe. We also found that the imputation error decreased as the number of hospital diagnoses and countries sampled increased, but that the rate of reduction in error shrank. Finally, we found that - given the existence of a reliable method for cost imputation - from a pure information standpoint, it is better to obtain estimates for fewer hospital diagnoses from more countries than the reverse.