Date Published: May 6, 2019
Publisher: Public Library of Science
Author(s): Pengsheng Ni, Molly Marino, Emily Dore, Lily Sonis, Colleen M. Ryan, Jeffrey C. Schneider, Alan M. Jette, Lewis E. Kazis, Dale W. Edgar.
This paper explores extreme response style to the Life Impact Burn Recovery Evaluation (LIBRE) Profile, a measure of social participation in burn survivors. We fit a Multidimensional Generalized Partial Credit Model (MGPCM) with a positive extreme response style (PERS) factor and compared this model with the original MGPCM, estimated the impact that PERS has on scores, and examined the personal characteristics that may result in an individual more likely to respond in a fashion that would inflate their true low scores. The average impact of the PERS, based upon the root mean squared bias, ranged from 0.27 to 0.50 of a standard deviation of the scale. Individuals who were older, had participated in a burn survivor support group, and had selected to self-administer the measure were less likely to have a high PERS bias that masks low scores. Future work can consider PERS when measuring the psychosocial impacts of burn injuries and other health conditions.
Clinicians increasingly use patient reported outcome measures (PROMs) to learn important information and optimize care.[1–3] However, PROMs can be sensitive to participants’ response styles, which may decrease the validity of the metric. One type of bias is extreme response style (ERS). A person who tends towards an ERS is more likely to select either the most negative or the most positive response to a question compared to an individual who does not have this bias with the same ability/true score. Individuals interpreting PROMs need to be aware of this phenomenon to account for this bias in results and better understand patient outcomes.
Burn survivors displayed PERS bias on three measures of social integration/participation that included the LIBRE Profile Family and Friends, Social Interactions, and Social Activities scales. The average impact of this bias, based upon RMSB, ranged from 0.27 to 0.50 of one standard deviation of the scale score.
To conclude, burn survivors are a clinically unique group of individuals. After finding evidence for a unique pattern of PERS in this group, it is possible that other unique clinical samples are likely to display PERS as well. When using self-reported measures in clinically unique samples, PERS should be estimated to assess if the bias is occurring and the extent to which it may be impacting scores. We examined the demographic characteristics for those with higher PERS scores but lower substantive factor scores. That information might be helpful for clinicians to interpret the scores for those subjects with characteristics that may impact scores. Clinicians might consider such information given that they have the required statistical support to identify respondents whose true scores may be lower than what they are reporting. Clinicians should be aware that PRO scores may be biased. In the case of PERS, this bias may conceal lower true scores for respondents that may impact provision of services when used as a needs assessment. When statistical support to clinicians is available, we recommend applying the multidimensional IRT model to estimate PERS scores and corresponding standard errors, and evaluate the impact of the PERS value by testing whether the PERS score is statistically significantly different from 0 (to do that, we will calculate the ratio between the PERS score and the standard error). This will provide the clinician with information on whether we can use the summary score or not. In the future, work can be done where such information is more easily available to clinicians through automated algorithms that are built where these computations are easily made with interpretative guidelines for their use.