OBJECTIVES: Accurately eliciting heath state utilities is at the cornerstone of effective patient-centered care. However, the initial data may have not been accurately elicited. Identifying data inconsistencies and minimally adjusting the initial health utilities, is a crucial step.

METHODS: It is assumed that a patient would feel worse when he/she experiences a number of adverse effects versus having only a subject of these adverse effects. This leads to the establishment of a monotonicity property in the values of the health state utilities related to combinations of adverse effects.

The proposed computational approaches seek to minimize the adjustment of the initial values and make them consistent in terms of this monotonicity property. A second source for inaccuracies is based on an assumption of independence on the way a patient reacts to various combinations of adverse effects. However, this independence assumption may be satisfied at various levels ranging from no independence to full independence. This case is also treated computationally as a minimization problem with a mechanism that controls the level of independence. Both methods are combined into a single approach that is based on quadratic optimization.

The proposed methods are illustrated on some published data related to non-metastatic prostate cancer.

RESULTS: Computational results indicate that the proposed optimization methods can control numerical weaknesses in initial data. At the same time, the degree of the inconsistencies can be quantified and it provides an additional window into the general state of the patient. The proposed methods are computationally practical and potentially very effective.

CONCLUSIONS: Health state utilities are often elicited under considerably strenuous conditions for the patient. Thus, they may be inconsistent. Being able to identify and deal with such inconsistencies is critical. The initial computational results are very encouraging and open new opportunities when using such data in medical applications.