An Item Response Theory Analysis of the Psychosis Screening Questionnaire in Ethiopia

Few psychosis screening instruments have been tested for use in Africa, yet appropriate tools can increase the detection of self-reported psychotic symptoms, improve the detection of psychosis and impact its prognosis.

The construct validity and factor structure of Psychosis Screening Questionnaire (PSQ) were tested with confirmatory factor analysis (CFA) and item response theory (IRT) in a sample of 1928 Ethiopian adults without any history of psychosis. We tested a unidimensional model with and without an item on mania. For IRT, unidimensional latent structure one-parameter logistic (1PL) and two-parameter (2PL) logistic models were tested and compared for relative fit using a likelihood-ratio test.

The prevalence of lifetime positive screens was 2.8% in an Ethiopian sample of adults from a general medical setting. A unidimensional model demonstrated good fit for the PSQ, (CFI = 0.993, TLI = 0.986 and RMSEA = 0.025). For IRT, a 2PL model was the best fitting one. IRT tests of item difficulty and discrimination parameters showed that paranoia had the highest discrimination $\left(\alpha =4.59\right)$ and lowest difficulty ( $\beta =2.53$ ), likely to be endorsed at low levels of psychotic features. Thought insertion had the highest item difficulty ( $\beta =2.98$ ). Overall, the measure captures the psychosis construct at higher levels of the latent trait and may be suited for detecting moderate to severe levels of psychosis.

The PSQ is found to have good construct validity in screening for psychosis among Ethiopian adults. Future studies may focus on the diagnostic validity of the PSQ comparing it with a structured clinical interview.