Split Choice

Split Choice in Clinical Practice

In a typical clinical setting, it may happen that one treatment has clear advantages over another, with side effects that are negligible in comparison with the potential gains. Most patients will accept such an intervention regardless of the strength of their preference for the likely health outcomes. On the other hand, there are many scenarios where some patients will decline an intervention that others will accept, simply because they place a lower relative value on its consequences. For some, the overwhelming concern may be a wish to avoid certain side effects, while others may simply attach less importance to the potential benefits of the treatment.

Examples include radical treatment for potentially fatal conditions (e.g., Stage I prostate cancer), prophylactic interventions in healthy individuals (e.g., mastectomy for women at high genetic risk of breast cancer), and prenatal testing for genetic abnormality. Many screening tests also fall into this category. In such cases, patients may make different treatment choices even though they have identical clinical prospects, with identical probabilities attached to the outcomes of their treatment. In other words, the clinical population splits into preference subgroups when confronted with the treatment choice. This can also apply when the effective treatment choice is deferred until after any side effects have come into play. Thus, a therapy with reversible side effects may be discontinued by some patients even though their objective clinical experience has been no worse than that of some who choose to continue.

Economic Evaluation and Split-Choice Bias

Economic evaluations for healthcare providers often quantify health benefits using preference-based comparisons of different health states. For example, the calculation of quality-adjusted life years (QALYs) is made by taking an average, over all health states, of the time spent in a state multiplied by a preference weight attached to that state. The preference weights are elicited as utility values between 0 (death) and 1 (the best conceivable state of health) from an appropriate population. In practice, a health state will generate different utility assessments from different individuals. Therefore an operational preference weight is obtained as a population-average utility.

This approach is unproblematic if all patients make the same decision when confronted with a treatment choice. The QALY value of a treatment [Page 1058]obtained from the average population utility will equate exactly to the QALY value that would be obtained if individual utilities were used instead: Subjects with higher and lower values will cancel each other out, since all receive the same treatment. In a split-choice scenario, the health benefit of a treatment can be experienced only by those who choose to accept it. It follows that the QALY valuation of a treatment ought to use a set of preference weights for the health state outcomes derived solely from the subpopulation of acceptors. This is an important stipulation, since it is plausible that treatment acceptors will attach higher utility values to the outcomes of a treatment than would those who choose to decline it. Indeed, a fully rational model for patient decision making would predict that those accepting treatment would do so just because of the higher expected utility that they associate with its consequences. Thus, a potential for bias arises in any economic evaluation in which preference weights for treatment outcomes are obtained from a population that includes individuals who would actually decline the treatment. Such split-choice bias can operate in one direction only, namely, to dilute the apparent effectiveness and hence the cost-effectiveness of the treatment in the group of patients who would choose to accept it. Moreover, the extent of the bias can be considerable and is sensitive to the degree to which individual decision making follows rational precepts. For example, it has been demonstrated that the QALY value could be underestimated by a factor of up to one half for treatments accepted by 70% of patients, with even larger biases possible at lower acceptance rates. Biases as large as this would lead to grossly distorted cost-effectiveness ratios.

...