Substituting proxy ratings for patient ratings in cancer clinical trials: An analysis based on a Southwest Oncology Group trial in patients with brain metastases

In studies of the effect of cancer treatment in the advanced disease settin g, researchers have attempted to avoid missing data for quality of life (QO L) assessments by either substituting proxy for patient assessments from th e outset or by interspersing proxy measures when patients are unable to res pond. Although poor agreement between patient and proxy assessments has bee n amply demonstrated in the literature, interest in using proxy measures pe rsists. Completion of the Spitzer QL-Index by a small sample of patients wi th brain metastases and family member proxies provided data for evaluating the ability to substitute proxy for patient QOL assessments. These data can not address treatment efficacy due to the modest sample size. Rather, the a nalyses serve to alert researchers to the important distinction (in a clini cal trial setting) between agreement and the use of the proxy as a surrogat e. We present several methods for evaluating the accuracy of proxy measures and for identifying other sources of error and bias that may vary with tim e or with treatment arm. Lin's concordance correlation coefficient suggests that proxies are generally a poor substitute for capturing a patient's per spective of his/her QOL. A longitudinal analysis suggests that the use of p roxy rather than patient responses could lead to different conclusions conc erning radiation therapy's effect on QOL.