Measuring Catastrophic Burdens with Household Survey Data: How Are Results Affected by the Length of Time Used to Measure Expenditures?

Presenter: Thomas M. Selden, Agency for Healthcare Research and Quality

Abstract

A growing literature examines the distribution of high or “catastrophic” out-of-pocket medical spending burdens. The literature examines the percentage of the population living in households with health care spending exceeding a threshold share of resources, thus providing a valuable perspective on the use of out-of-pocket payments to finance an inherently high-variance good such as health care. Catastrophic burdens have been studied in a growing number of developed and developing economies. Not all countries, however, have data that are equally well suited for burden analyses.

Of particular interest for our paper are the potential pitfalls associated with measuring burdens with data on health spending that cover only a narrow window of a single month or less. An important example are the Living Standards Measurement Survey data, which are commonly used for burden analyses in developing countries. Health care spending can be highly episodic in nature. Recent evidence from the U.S. shows that nearly half of all family spending occurs in a single month of the year (Selden, 2008). Clearly, data for a single month of spending can yield burden measures that differ greatly from what one would find with a full year of expenditure data. How comparable, then, are burden estimates across countries with varying data durations? How do differences in the time period studied affect overall burden prevalence, the regressivity of burdens, or their incidence among subgroups such as the chronically ill?

Our study uses U.S. data from the Medical Expenditure Panel Survey (MEPS) to provide a partial answer to these questions. A key advantage of MEPS for this analysis is that it provides data on the timing of events. This, plus detailed data on employment, permits the calculation of burdens using both annual and single months, thereby offering a direct comparison of the two approaches using a single dataset.

Preliminary results suggest that burden prevalence is remarkable stable to the duration over which expenditures are measured. So, too, is the distribution of burdens across income deciles. The paper examines these results further to assess differences across subgroups defined by insurance status, race/ethnicity, age, and the presence or absence of a variety of chronic conditions.