Comparison of Patient and Proxy Responses to the Stroke Impact Scale
Purpose & Methods: Our purpose was to compare patient and proxy responses to the Stroke Impact Scale ( SIS). We recruited 150 patients (mean age = 73±10) and their proxies (mean age = 59±15) from the Kansas City Stroke Registry at 90 to 120 days post stroke. Fifty-one percent of patients were male and 75% of proxies were female. The precision with which proxy responses replicated patient responses was estimated using intraclass correlation coefficient (ICC) obtained from random effects ANOVA model, while bias and precision of the proxy response as an estimate of patient response were assessed with linear models. Results: Two SIS subscales (ADL/IADL and hand function) and the aggregate physical function subscale showed no evidence of bias. Remaining subscales showed a consistent pattern of proxy overestimating patient scores at the low end of the scale and underestimating at the high end. However, with the exception of the emotion and communication scales (which are biased high over most of the range), the biases appear to be mathematical artifacts created by the bounds of the scales. Precision was reasonable for strength (.56), emotion (.50), communication (.42), and memory/ thinking (.43). ICC s were higher for social participation (.59), mobility (.66), ADL/IADL (.65), hand function (.81), and the aggregate physical domain (.78). Conclusion: In observable behavior (e.g. ADL/IADL, mobility, hand function, and combined physical functioning), our data indicate that proxies can be used as surrogates in clinical studies. Another important implication of this data is that clinical researchers must not develop one summary score of different health related domains.