Response shift in quality of life ratings in homeless individuals wih mental illness: a residuals analysis of the at home/chez soi study

Background: The At Home/Chez Soi project was a pan-Canadian randomized controlled trial of a Housing First intervention among 2,148 homeless individuals with mental illness. The trial provided subsidized private-market apartments with recovery-oriented support services to treatment (HF) participants, while Treatment As Usual (TAU) controls were left to obtain services available in the community. Although the HF group reported significant improvements in quality of life (QOL) compared to controls, greater improvements were expected given observed group contrasts in both housing stability and a subsample's qualitative interviews. Recognized in the literature as a bias affecting self-reported QOL measurements, response shift occurs when a life event (e.g., an illness or a treatment) changes the meaning of an individual’s QOL rating. Their internal standards may change or different values defining QOL may be reprioritized or redefined, affecting the accuracy of longitudinal quality of life comparisons, such as those of the At Home/Chez Soi project. Following reviews of the literature on QOL among individuals with mental illness and among the homeless, as well as a review of theories of response shift and methods for its identification, the current thesis seeks to adjust for response shift in comparisons of QOL in the At Home/Chez Soi participants.Methods: A secondary analysis method developed in previous literature allows for the identification of response shift using the residuals from an explanatory model of the QOL outcome. Based on a random intercept model explaining variability in the 7-point global QOL item (of the 20-item QOL index [QOLI-20]), participants’ degree of response shift was defined by the standard deviation (sd) of their respective residual values. These sd values were accounted for in group comparisons of both the QOLI-20 total score and global item outcomes, allowing for estimates of the intervention’s effect on QOL to vary according to degree of participants’ response shift. A sensitivity analysis reassessed findings after excluding observations for which interviewers felt responses were invalid or insincere, under two levels of exclusion criteria (A and B).Results: A model explaining 62% of QOL variability was estimated using 23 covariates. Participants’ mean sd of residuals was 0.95 (min: 0.00, max: 3.93). HF treatments effects on QOLI-20 total scores diminished when estimated at higher levels of response shift (1 point increments in sd of residuals), for most follow-up periods (interaction at 6 mo. β = -3.97, p = .052; at 12 mo. β = -3.20, p = .109; 18 mo. β = -4.15, p = .043) but not at 24 months (β = 0.14, p = .946). A non-significant decrease was seen in the HF treatment odds ratio (OR) from an ordinal logistic regression of the QOLI-20 global item, collapsed across follow-up periods (interaction β = 0.83, p = .175). Following sensitivity analyses, the HF treatment x sd of residuals interactions no longer met significance levels as exclusion criteria were applied to the QOLI-20 total score analyses. However, the interaction term of the QOLI-20 global item analysis became significant under the more stringent exclusion criterion B (β = 0.69, p = 0.027), while criterion A had little impact on the original estimate. Conclusion: These findings suggest that unexpectedly modest effects of the HF intervention on QOL may be explained by response shift, a novel observation in the homelessness literature. A more conservative conclusion from the results is that QOL improvements from a HF intervention are more evident in individuals for whom reported and expected QOL differ by an amount that is relatively consistent over time. HF treatment effects are smaller in participants whose residual patterns fluctuate over time. This fluctuation is interpreted as response shift, in accordance with previous literature, though alternative interpretations may also explain some of these patterns.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.