monash_2604.pdf (4.16 kB)
Negative Utility Scores and Evaluating the AQoL All Worst Health State
journal contributionposted on 08.06.2017, 02:31 by Richardson, Jeff, Hawthorne, Graeme
This paper is concerned with two issues. One is a general problem for the evaluation of very poor health states and the other, while arising from the scaling of the Assessment of Quality of Life (AQoL) instrument is a more general problem associated with the use of multiplicative multi-attribute instruments. These problems are: (i) the treatment of negative utility values; (ii) the utility score on a `life-death scale' of the `all worst' health state described by an instrument (in this particular paper in reference to the AQoL, although the issue is pertinent to all such instruments). The first problem arises from the fact that there is no lower limit to the negative utility scores implied by responses in a conventional time trade-off (TTO) interview. A respondent who indicates that they would not accept any time in a health state worse than death, even when this was followed by full health, implies a utility score of minus infinity for this health state. While not discussed in this paper, the same outcome is obtained from the standard gamble when a respondent refuses to contemplate a health state at any finite probability. A score of minus infinity or even a very large negative utility score has no meaning and such responses must be transformed into a lower, albeit negative, score. The second problem is that the disutility values generated by a multiplicative model - as used in the AQoL - vary from 0.00 to 1.00. These `model utilities' - utilities measured in `model space' - must be rescaled so that they represent utilities on a life-death scale where, following convention, a utility of 1.00 corresponds with `full' health, and a utility score of 0.00 corresponds with death. In principle it is easy to rescale the model scores. `Full health' has a common numerical value - 1.00 - on both the model and the life-death scales. The transformation then only requires information on the correspondence between one other point on the two scales. The simplest such point is the instrument `all worst' (the health state described when each item of the instrument is at its worst level). However scaling this point may require respondents to visualize a particularly complex and unusual state. For example, the AQoL `all worst' is a health state with 12 dimensions. However, the raison d'être of the decomposed, multi-attribute approach to health state measurement is the avoidance of the need to carry out such a cognitively complex task. The likelihood of error is further increased when the instrument's all worst health state is used to establish the nexus, if - as with the AQoL - the all worst health state is close to death and respondents have never experienced a health state so bad that death was equally (un)desirable. Procedures adopted for the scaling of the AQoL are outlined and discussed. It is concluded that both of the above problems are quantitatively significant and have received too little discussion in the literature. Readers wishing to understand the scoring of the AQoL are referred to the companion paper, Utility Weights for the 'Assessment of Quality of Life' (AQoL) Instrument (Hawthorne, Richardson et al. 2001). Details can also be found in the AQoL user manual, Using the Assessment of Quality of Life (AQoL) Instrument (Hawthorne, Richardson et al. 2000).