Using the range of cell mates’ experience this year as the distribution of possible expenditures next year for each individual or family in the cell permits the calculation of distributions with only 1 year of cross-sectional data. Cells would be defined by families’ characteristics as close to the start of the year as data permit so that the rest of the year is prospective to those definitions. Individuals could be grouped into cells by predicted next-year expenses. Handel (2011), for example, uses adjusted clinical groups (a case-mix system based on claims-defined diagnoses) together with other characteristics to do this, but the diagnosis cost groups form of risk adjustment system, RxRisk (a risk assessment instrument that uses automated pharmacy data systems to characterize chronic conditions to predict future costs), or some combination of relative risk algorithms could also be used (see the description of methods for the Dutch health insurance system in van de Ven et al., 2007, for a mixed risk equalization/adjustment system). Either total expenditures or discrete types of expenditures could be grouped.7 The Gaussian copula methods used by Handel to combine different types of spending could be adapted to account for within-family correlation in grouping individual out-of-pocket medical care spending into a family-level variable.
To estimate the full range of spending next year from this year’s data, one must adjust for the general increase of spending to be expected next year, and for any people missing from the data. If the survey does not include people who died or entered institutional care during the year, their numbers and risk of spending experience could be estimated separately based on cell characteristics and the outcomes for these virtual people (who are missing at the follow-up survey) added to the range of possibilities in the cell for those who survived or who are in a noninstitutional setting by the end of the year. One could either base the cells on out-of-pocket spending or base them on total spending, and then use a standard insurance policy or actual terms of coverage to calculate the resulting out-of-pocket spending.
Each member of the cell would be assumed to have the same distribution, which should be acceptable for combining with their family resources and thresholds to calculate percentiles in the tail of the distribution that
7 In the absence of detailed information on different coverage for different types of health care services, it may be sufficient to examine out-of-pocket expenditures combining all of the types of care or to group classes of insurance into the following categories: uninsured, public based on poverty or categorical eligibility, Medicare, group insurance. We are not aware of any major national data set that contains the level of detail on coverage that Handel (2011) has and that also has an adequate response rate and spans the age range necessary for this task. See the work of Goldman and colleagues on the Future Elderly Model, a demographic and economic simulation model designed to predict the future costs and health status of the elderly, at http://roybal.healthpolicy.usc.edu/projects/fem.html.