Framing mental health questions using context better predicts disability than does frequency

Most mental health questionnaires only probe behavior and emotion over one domain (e.g., frequency) by using an adverbial phrase such as 'How often...?'. This single-domain approach to behavior sampling may limit the scope of psychiatric evidence that these instruments can capture. We investigated whether adding domains such as intensity, duration, ability and context to mental health survey instruments garner responses that better characterize or measure relevant symptoms or behaviors, testing its association with relevant outcomes. We tested whether our multidomain instrument could be associated with hospitalization, self-harm, education, and disability. 1,504 participants took an augmented version of the SWAN questionnaire (‘SWAN+’), where 4 domains were added to the ability domain already present (frequency, duration, intensity, and context). First, we used structural equation models and the tri-factor modelling to analyze if the latent construct of domains was independently associated with the outcomes. Our results suggest that adding domains such as frequency and context could help characterize disability beyond the domain originally probed by the SWAN questionnaire. Different domains had different predictive effects across outcomes measured in the study. We also complemented results from the tri-factor modelling with a more classical approach, by performing regressions on the summed scores on each domain and the entire scale, adjusted by age and gender. Second, analyses were performed to identify the best predictor domain within each question to predict disability. Significant results were found for the ability and context domains on practically all questions (p<0.001), revealing that they are important for almost all questions, whereas frequency is predictive for only a subset of them. This confirmed the results of the tri-factor analysis demonstrating the clinical utility of our findings. It also indicated that context is more predictive than frequency, that ability and context should be uniformly applied, and that the ubiquitous frequency phrasing is not uniformly as informative in our study.Overall, our results suggest that the ‘multidomain’ approach to questionnaires may improve inferences about mental health and predictions of functional outcomes. Framing questions using appropriate and relevant domains seems critical to measure psychopathology adequately, as opposed to asking solely about the frequency or recurrence of any given symptom.