Experimental Quality

Measuring health utilities: Why sequencing matters

Let’s say I ask you–on a scale from 0 to 1–what your utility would be if you had cancer. This is a difficult question for most people to answer not only because many people will not have cancer and will not know what it is actually like, but also the concept of “utility” is not well understood.

Another approach to measure utility is to use a time trade-off (TTO) approach. One could ask, would you rather live 10 years with cancer or X years in perfect health. If you think that 5 years in perfect health is the same value as 10 years with cancer, your utility value would be 0.5.

While this approach appears to solve the direct elicitation problem, what a paper by Pinto-Prades et al. (2019) finds is that when you answer multiple questions, your answer to previous questions will influence the answer to this question.

Let’s say I asked you a TTO question about being paralyzed and in constant pain–rather than having cancer–and your answer was 4 years, for a utility of 0.4. If I think asked you about having cancer, you may think, “well, being paralyzed and in pain is way worse than cancer, so I’ll answer 6 years.” Thus, your previous utility estimate of 0.5 went up to 0.6. On the other hand, if I began by asking the TTO question about a paper cut, you might say 9.9 years (or a utility of 0.99). If I then ask about cancer, you may think “well, having cancer is way worse than a paper cut, so I’ll answer 4 years.” Thus, now the lung cancer utility has decreased to 0.4.

The paper by Pinto-Prades notes that there are two reasons why sequences matter: context effects and preference imprecision. First, is that people are better at rank ordering health states than measuring quality of life based on a numerical number. It is likely that I know which health state I prefer, and maybe even by how much, but the absolute level of preference is difficult to quantify. Second, there is some imprecision in individual preferences. Thus, consider the case where I know that (i) being paralyzed has a utility between 0.35 and 0.45, (ii) having cancer gives a utility between 0.40 and 0.50, and (iii) I know that I would rather have cancer than be paralyzed. If I answer 0.35 for being paralyzed, the cancer value could be anywhere between 0.40 and 0.50. If I answer 0.45, however I know that the lung cancer value now must only be between 0.46 and 0.50, so I do not violate dominance. Thus, my answer to previous questions will inform my current answer.

Pinto-Prades and co-authors provide empirical evidence of this phenomenon. Using an EQ‐5D‐3L-based survey of nearly 1500 respondents, the authors tested whether question sequence affected the utility estimates (also controlling for patient characteristics). They find that:

…the utility of the Best (Worst) health state…is higher (lower) when it appears in an Ascending (Descending) sequence than in a mixed sequence… The utility of the intermediate health state is also statistically significantly lower (higher) when it follows the Best (Worst) health state than when its evaluated first in the sequence…

In summary, when measuring health utilities using TTO, be sure to randomize question order to prefer sequences from biasing your results.


Leave a Reply

Your email address will not be published. Required fields are marked *