Background: The finding of variation in the optimal threshold of the General Health Questionnaire (GHQ) across different settings has proved difficult to explain. This analysis aimed to examine the optimal threshold of the GHQ, its variability and relationship with prevalence of psychiatric disorder.
Methods: A cross-sectional two-phase epidemiological survey was used in a study of non-psychotic psychiatric disorder of General Practice consulters. A total of 1670 consecutive patients were screened using the GHQ-28 and the GP encounter form, and 336 were interviewed using the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID). The total prevalence of the psychiatric disorders was estimated using three different methods and was calculated in the rural and urban practices and among those attending the General Practitioners and Practice Nurses.
Results: The frequency distribution of the GHQ score for the whole sample of respondents was skewed resulting in the biased mean GHQ score. The mean values for different sample categories were found to be higher than the median values indicating that the median GHQ score may be a better parameter to describe the score distribution than the mean. The median value of the GHQ score showed a strong correlation with the prevalence estimated by the three different methods.
Conclusions: The median of GHQ score gives a better estimate for the cut-off score than the mean. The median score may be used as an estimate for the optimal threshold, in settings where the sensitivity and specificity are not known or where an estimate of the prevalence from a questionnaire survey is required. Alternative methods are preferable when screening for individual cases or in the context of a two-phase design.