Comments

Health Affairs encourages readers to engage in discussion via comments on our Web site.

• To RESPOND to a particular article: Click on the link "Submit a response to this article" in the box at the top right-hand corner of the article.
• To READ responses to a particular article: Click on the link "View responses" in the box at the top right-hand corner of the article.

Comments to:

Kenneth E. Thorpe, Curtis S. Florence, David H. Howard, and Peter Joski, The Rising Prevalence Of Treated Disease: Effects On Private Health Insurance Spending, Health Affairs Web Exclusive, June 27, 2005 [Abstract] [PDF] [HTML Version][Interview] [Reprints & Permissions]

Comments:Submit a response to this article

Comments published:

Thorpe Analysis Of Reasons For Health Care Cost Increases Might Be Wrong

Regina E. Herzlinger   ( 20 December 2005 )

Different Approach, Same Conclusion

Kenneth E. Thorpe, Curtis S. Florence, David Howard, Peter Joski   ( 22 December 2005 )

Thorpe Analysis Of Reasons For Health Care Cost Increases Might Be Wrong 20 December 2005
Regina E. Herzlinger, Nancy R. McPherson Professor Harvard Business School Send comment to journal: Re: Thorpe Analysis Of Reasons For Health Care Cost Increases Might Be Wrong rherzlinger{at}hbs.edu Regina E. Herzlinger	Ken Thorpe and his colleagues attribute the cost increases for certain diseases primarily to increases in prevalence; but their analysis does not appear to support their conclusion. The point might appear obscure, but it is important to get this right, because the assertion that health care cost increases are caused primarily by use, or prevalence, rather than price per case can be used to justify rationing or other policy prescriptions that restrict the provision of health care, as Thorpe has done elsewhere. (See, for example, Kenneth E. Thorpe, The Rise In Health Care Spending And What To Do About It, Health Affairs 24, no. 6, 1436-1445.) The authors attribute the change in total cost to one of two causes: changes in volume (prevalence) or to cost per case. But both calculus (remember those partial derivatives?) and geometry (see my graphical example, at http://www.people.hbs.edu/mlaisne/Hyperlipidemia.jpg) tell us that these changes must be attributed to three causes: volume, cost per case, and the joint effect of changes in volume and cost per case. To reach their conclusion, Thorpe et al have attributed the joint effect to prevalence; however, this joint variance is joint -- i.e., it cannot be clearly attributed, absent further econometric analysis, to changes in either cost or prevalence. This point could alter their conclusions. For example, my analysis of their numbers for hyperlipidemia (see their Exhibit 1) shows the cost variance to be approximately $472; the prevalence variance, $1,680; and the joint variance, $2,054--all in thousands per 100,000. The total difference between the 1987 and 2002 costs of the disease of $4,205 is thus 11% explained by changes in cost per treated case, 40% by changes in prevalence, and 49% by changes in both. Thorpe et al. state that 89% of the difference in total cost of the disease is explained by changes in prevalence by adding the prevalence variance (40%) and the joint variance (49%), but if the joint variance were attributed to changes in the cost, the percentage explained by changes in price would rise to 60%--a conclusion directly opposite to theirs. Neither conclusion, however, is warranted by the analysis. It can be used only to differentiate the separate impact of the cost, prevalence, and joint variances. I have used the Thorpe et al. paper in my own work, and others might have found it important as well. It would be useful if they either responded to this view of their conclusions in the two articles that I have read on this topic in Health Affairs -- there may be more, or they may have published them elsewhere too -- or else retract them, redo the analysis, and have them published again.
Different Approach, Same Conclusion 22 December 2005
Kenneth E. Thorpe, Professor Emory University, Curtis S. Florence, David Howard, Peter Joski Send comment to journal: Re: Different Approach, Same Conclusion kthorpe{at}sph.emory.edu Kenneth E. Thorpe, et al.	We read with interest Regina Herzlinger's eLetter asking for clarification of the methods used in our recent Web Exclusive article on the prevalence of treated disease. In this eLetter, she makes three assertions regarding our paper. The first is that our methods systematically overestimate the impact of increases in treated prevalence on health care spending. The second is that our general conclusion regarding the qualitative importance of the impact of increases in treated prevalence on health care spending is invalid. The final assertion is that our results have been used by the lead author to advocate nonprice rationing of health care services. In this response we describe why our calculations differ from Herzlinger's and show that her assertions are incorrect. Our decomposition of total spending increases into components attributable to increases in treated prevalence (TPV) and spending per case (SPC) is based on the following formula. First, start with the difference in total spending: TPV2002 × SPC2002 – TPV1987 × SPC1987. Next, add and subtract TPV2002 × SPCSPC1987. Rearranging terms yields the amount attributable to increases in treated prevalence, TPV2002 × SPC1987 – TPV1987 × SPC1987, and the amount attributable to increases in spending per case, TPV2002 × SPC2002 – TPV2002 × SPC1987. Adding these two terms together equals the change in total spending for the condition. This is a straightforward accounting identity. This form of the decomposition implicitly assumes that changes in spending per case and treated prevalence are independent. Alternatively, one could add and subtract a few terms to the change in total spending, as Herzlinger has done, so that the change in spending is (SPC2002 – SPC1987) × TPV1987 + (TPV2002 – TPV1987) × SPC1987 + (SPC2002 – SPC1987) × (TPV2002 – TPV1987). This formula is algebraically equivalent to the one we used (they both add up to the change in total spending), but in this case the change in total spending is decomposed into changes in spending per case, changes in treated prevalence, and changes in spending per case and treated prevalence jointly. The degree to which these two formulations produce different results depends on the joint variance of spending per case and treated prevalence. Determining whether accounting for the joint variance renders our conclusions invalid requires performing these calculations for all conditions. Herzlinger's main assertion, that our methods attribute any joint variance between spending per case and treated prevalence to treated prevalence, is incorrect. We used her approach across all twenty conditions, and in some cases, her calculation produces even higher estimates for the percentage change in spending associated with treated prevalence than we estimated in our paper (for example, mental disorders, back problems, and cerebrovascular disorders, among others). Even if one undertook the analysis Herzlinger suggests, treated prevalence still accounts for most of the rise in spending across the twenty conditions we examined (the average percentage increase linked to treated prevalence is 70 percent across all twenty conditions), even without allocating any of the joint variance to increases in prevalence or spending per case. Since the percentage change in prevalence exceeded the percentage change in spending per case for thirteen conditions, we believe that a reasonable allocation using her approach reinforces our results and conclusions. However, rather than looking across the twenty conditions, she produced an example using one condition — hyperlipidemia — that resulted in the largest (49 percent) joint variance (residual), with 40 percent of increased spending associated with increased treated prevalence and 11 percent associated with increased spending per case. In this case, however, treated prevalence increased 437 percent, compared with a rise in spending per case of 122 percent between 1987 and 2002. We believe that any reasonable method for allocating the joint variance would apportion most of it in the treated prevalence column, leaving our conclusion for this condition unaffected. Finally, Herzlinger contends that the results have been used by one author (Thorpe) to justify rationing. This is also incorrect. The Thorpe article in question focuses on the use of population-based health promotion and wellness strategies designed to reduce the rise in obesity to address the long-term rise in spending.