Measurement properties of PROMIS Sleep Disturbance short forms in a large, ethnically diverse cancer cohort

AIMS: To evaluate model fit, differential item function (DIF), and construct validity of select short forms from the PROMIS Sleep Disturbance item bank. METHODS: We recruited cancer survivors who were between 6 13 months post diagnosis (n = 4,956), as part of the Measuring Your Health (MY-Health) study. We measured sleep disturbance using 10 items commonly found in PROMIS Sleep Disturbance short forms (Sleep 4a, Sleep 6a, Sleep 8b), and which are frequently administered in computerized adaptive testing. We evaluated domain reliability using Cronbach’s coefficient alpha and factorial validity by fitting a PROMIS Sleep Disturbance unidimensional measurement model using confirmatory factor analysis (CFA). At the item-level, we examined DIF with respect to race/ethnicity (non-Hispanic White [NHW], non-Hispanic Black [NHB], Hispanic, and Asian/Pacific Islander), age, and sex. We used a multi-group CFA and multiple indicators, multiple methods (MIMIC) analyses. We then assessed construct validity (convergent, discriminate, and known groups) for sleep short forms, and a new “best fit” 6-item sleep disturbance short form. 1 Correspondence concerning this article should be addressed to: Roxanne E. Jensen, Ph.D., Cancer Prevention and Control Program, Lombardi Comprehensive Cancer Center, Georgetown University, 3300 Whitehaven Street NW, Suite 4100, Washington, DC 20007, USA; e-mail: [email protected] 2 Department of Oncology, Georgetown University, Washington DC, USA 3 Pharmerit International, Boston, MA, USA 4 Royal College of Surgeons in Ireland, Dublin, Ireland 5 Lineberger Comprehensive Cancer Center, University of North Carolina, Chapel Hill, NC, USA 6 Department of Health Policy and Management, University of North Carolina, Chapel Hill, NC, USA 7 Public Health Sciences Division, Fred Hutchinson Cancer Research Center, Seattle, WA, USA 8 Cancer Prevention and Control Program, Lombardi Comprehensive Cancer Center, Washington 9 Columbia University Stroud Center and New York State Psychiatric Institute, New York, NY, USA 10 Research Division, Hebrew Home at Riverdale, RiverSpring Health, Riverdale, NY, USA R. E. Jensen et al. 354 RESULTS: We identified a satisfactory unidimensional sleep disturbance 6-item measure (χ(6)37.6, p < 0.001, RMSEA = 0.031). To achieve this, we removed four items from the model with item content overlap and added residual covariances between positively worded items in order to address a method effect. We identified one instance of DIF: NHW participants were less likely to agree with the statement “I had difficulty falling asleep” compared to NHBs, Hispanics, or Asians/Pacific Islanders, who all reported the same level of sleep disturbance. After controlling for DIF, we extended this into a MIMIC model, identifying no additional DIF by age or sex. Across all race/ethnicity groups, the adjusted overall means suggest that older adults reported significantly lower sleep disturbance, and NHW, NHB, and Hispanic women reported significantly higher sleep disturbance than male survivors of the same race/ethnicity. CONCLUSIONS: We could not fit a unidimensional measurement model for either the full 10items, or for any combination of sleep disturbance items used in PROMIS Sleep Disturbance short forms. However, after we removed the overlapping item content and adjusted for methods effects, a 6-item measurement model for sleep disturbance fit the data well, with very little evidence of substantial DIF. This suggests this new measure (Sleep 6b) can be used in different groups across the adult lifespan, and in males and females in a heterogeneous cancer population. Our findings suggest further validation work is necessary to understand the impact of reverse-scored items, response set effects, and content overlap in this item bank.