that the most predictive of variables include age, gender, and health status (status per se or case-mix and severity). But in moving to measures of the family distribution of out-of-pocket spending or its variance, how does one combine individual data into a meaningful model for the family? There is very little guidance on this score in the literature.

The panel thinks that the U.S. Department of Health and Human Services needs to consider several possible alternative analyses to help better understand these issues. The Agency for Healthcare Research and Quality, the Office of the Assistant Secretary for Planning and Evaluation, or both agencies need to perform a series of analytical studies using MEPS. The results of these analyses can be used to inform the move from a purely retrospective approach that uses medical care economic burden as a proxy for risk to an approach that estimates risk directly.

These studies should include an analysis of both the cell-based approach to estimating the expected amount of spending and the use of regression methods to understand the expected risks; both are important to the development of appropriate alternatives to the short-term strategy that we have offered. The needed analyses should address the following questions:

  • At the family level, how does current out-of-pocket medical care spending or its two major components (out-of-pocket premiums and other out-of-pocket payments for services) predict next year’s out-of-pocket spending? How stable is this relationship in the near term?
  • If one expands the covariate list to include other family characteristics besides the first-year out-of-pocket medical care spending, what relationships can be seen in terms of predictive ability?
  • Because individual characteristics are the strongest predictors of future average expenditures, how does one roll up individual predictions into a composite family measure that is predictive of future family out-of-pocket medical care expenditures? If there is an indicator for having a chronic disease or cluster of diseases for any member of the family, how well does this (and similar family constructs based on individual health characteristics) predict future out-of-pocket spending in the prospective measure or explain burden in the retrospective measure?
  • If one begins with the distribution of individual expenditure distributions net of observed individual and family characteristics, how does one best combine these into the family’s distribution around its expected amount? The Handel (2011) approach has some promise, but one needs to know how well that approach actually approximates the family’s distribution, especially in the


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement