Background: No convincing explanation has been forthcoming for the variation in best threshold to adopt for the GHQ in different settings.
Methods: Data dealing with the GHQ and the CIDI in 15 cities from a recent WHO study was subjected to further analysis.
Results: The mean number of CIDI symptoms for those with single diagnoses, or those with multiple diagnoses, does not vary between cities. However, the best threshold is found to be related to the prevalence both of single and of multiple diagnoses in a centre. Variations in the diagnoses to be included in the 'gold standard' did not account for the variation observed. There was a strong relationship between area under the ROC curve (as a measure of the discriminatory power of the GHQ) and the best threshold, with higher thresholds being associated with superior performance of the GHQ. The items on the GHQ-12 that provided most discrimination between cases and non-cases varied from one centre to another.
Conclusions: The GHQ threshold is partly determined by the prevalence of multiple diagnoses, with higher thresholds being associated by higher rates of both single and multiple diagnosis. The mean GHQ score for the whole population of respondents provides a rough guide to the best threshold. In those centres where the discriminatory power of the GHQ is lowest, it is necessary to use a low threshold as a way of ensuring that sensitivity is protected, but the positive predictive value of the GHQ is then lower. Some of the variation between centres is due to variation in the discriminatory power of different items.