Purpose: To validate the underlying mathematical process of Goal Attainment Scaling (GAS), as an illustration of the problems encountered by patient-generated indexes in the quest of patient-centred outcomes.
Methods: Data are simulated to represent the type of input to GAS. Rasch analysis is used to linearize the response categories for each variable associated with each goal, thus making it possible to compare the ordinal non-linear outcome of the GAS process with its linear equivalent, under the assumption of strict unidimensionality. Using a minimum clinically important difference (MCID), the level of difference between the two estimates is assessed.
Results: Over 14% of the simulated cases showed a magnitude of difference in change scores between the ordinal and linear-based GAS scores greater than the MCID. These differences were most likely to occur when patients start or finish their GAS scores at the margins of the score range, where non-linearity is greatest. The results show that the GAS process does not support mathematical operations such as multiplication. Apparent clinically meaningful changes scores can be generated solely from the non-linear nature of ordinal scores.
Conclusions: Using patient-centred approaches to measurements such as GAS presents formidable scientific challenges. Suggestions are made which, in the context of GAS applications, may overcome some of these limitations. This involves the establishment of 'item banks' of goals which can be pre-calibrated onto a unidimensional metric such that linearized versions of the various scores (e.g., difficulty) could be imported into the process.